Hugging Face's logo

yamahei21
/
huihui-qwen3.5-9b-int4-rht-rtn

Safetensors
uzu
quantization
int4
rht
8-bit precision
Model card Files
xet
 Community
yamahei21 commited on
Commit
3c6d7ab
·
verified ·
1 Parent(s): 052303b

Add 9B RHT+INT4 RTN checkpoint

Browse files
Files changed (7) hide show
  1. .gitattributes +1 -0
  2. README.md +37 -0
  3. config.json +2127 -0
  4. encoding.json +4 -0
  5. model.safetensors +3 -0
  6. provenance.json +11 -0
  7. tokenizer.json +3 -0
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Huihui Qwen3.5-9B INT4 (RHT + Group32 RTN)
2
+
3
+ Experimental uzu/lalamo checkpoint. **RTN only**; YAQA/QAD not applied yet.
4
+
5
+ ## Base model
6
+ - `huihui-ai/Huihui-Qwen3.5-9B-Claude-4.6-Opus-abliterated`
7
+ - untied embedding/lm_head, 32 layers, hidden 4096
8
+
9
+ ## Quantization
10
+ - Linear: RHT (Hadamard block 32, input_output) + asymmetric INT4 group 32, layout output_input
11
+ - Input embedding: INT4 (output-only RHT)
12
+ - lm_head/readout: INT4 (input-only RHT)
13
+ - Norm: BF16/F32 as stored
14
+ - lalamo commit: `e401e17761f2fcd337b0cf20218d67d656e6ecba`
15
+ - uzu commit: `fa0aa23d`
16
+
17
+ ## Size
18
+ - `model.safetensors`: ~5.19 GB
19
+
20
+ ## uzu usage
21
+ ```bash
22
+ hf download <repo> --local-dir huihui-9b-int4-rht-rtn
23
+ uzu cli --model huihui-9b-int4-rht-rtn -m "Say hi in one word."
24
+ ```
25
+
26
+ ## Measured on M3 16GB
27
+ - EN: TTFT 0.26s / prefill 68.8 t/s / gen 6.28 t/s / 4.95GB
28
+ - JA: TTFT 0.27s / prefill 63.0 t/s / gen 11.47 t/s / 4.96GB
29
+
30
+ ## Status / limitations
31
+ - Experimental. Not instruction-tuned after quantization.
32
+ - No YAQA/QAD yet.
33
+ - JA output has factual errors from the base model, not necessarily quantization.
34
+ - License follows the base model and Qwen3.5 terms.
35
+
36
+ ## Provenance
37
+ See `provenance.json`.
config.json ADDED
@@ -0,0 +1,2127 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "LanguageModelConfig",
3
+ "token_codec_config": {
4
+ "type": "ChatCodecConfig",
5
+ "prompt_template": "{%- if tools %}\n {{- '<|im_start|>system\\n' }}\n {%- if messages[0].role == 'system' %}\n {{- messages[0].content + '\\n\\n' }}\n {%- endif %}\n {{- \"# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou are provided with function signatures within <tools></tools> XML tags:\\n<tools>\" }}\n {%- for tool in tools %}\n {{- \"\\n\" }}\n {{- tool | tojson }}\n {%- endfor %}\n {{- \"\\n</tools>\\n\\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\\n<tool_call>\\n{\\\"name\\\": <function-name>, \\\"arguments\\\": <args-json-object>}\\n</tool_call><|im_end|>\\n\" }}\n{%- else %}\n {%- if messages[0].role == 'system' %}\n {{- '<|im_start|>system\\n' + messages[0].content + '<|im_end|>\\n' }}\n {%- endif %}\n{%- endif %}\n{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}\n{%- for message in messages[::-1] %}\n {%- set index = (messages|length - 1) - loop.index0 %}\n {%- if ns.multi_step_tool and message.role == \"user\" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}\n {%- set ns.multi_step_tool = false %}\n {%- set ns.last_query_index = index %}\n {%- endif %}\n{%- endfor %}\n{%- for message in messages %}\n {%- if message.content is string %}\n {%- set content = message.content %}\n {%- else %}\n {%- set content = '' %}\n {%- endif %}\n {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) %}\n {{- '<|im_start|>' + message.role + '\\n' + content + '<|im_end|>' + '\\n' }}\n {%- elif message.role == \"assistant\" %}\n {%- set reasoning_content = '' %}\n {%- if message.reasoning_content is string %}\n {%- set reasoning_content = message.reasoning_content %}\n {%- else %}\n {%- if '</think>' in content %}\n {%- set reasoning_content = content.split('</think>')[0].rstrip('\\n').split('<think>')[-1].lstrip('\\n') %}\n {%- set content = content.split('</think>')[-1].lstrip('\\n') %}\n {%- endif %}\n {%- endif %}\n {%- if loop.index0 > ns.last_query_index %}\n {%- if loop.last or (not loop.last and reasoning_content) %}\n {{- '<|im_start|>' + message.role + '\\n<think>\\n' + reasoning_content.strip('\\n') + '\\n</think>\\n\\n' + content.lstrip('\\n') }}\n {%- else %}\n {{- '<|im_start|>' + message.role + '\\n' + content }}\n {%- endif %}\n {%- else %}\n {{- '<|im_start|>' + message.role + '\\n' + content }}\n {%- endif %}\n {%- if message.tool_calls %}\n {%- for tool_call in message.tool_calls %}\n {%- if (loop.first and content) or (not loop.first) %}\n {{- '\\n' }}\n {%- endif %}\n {%- if tool_call.function %}\n {%- set tool_call = tool_call.function %}\n {%- endif %}\n {{- '<tool_call>\\n{\"name\": \"' }}\n {{- tool_call.name }}\n {{- '\", \"arguments\": ' }}\n {%- if tool_call.arguments is string %}\n {{- tool_call.arguments }}\n {%- else %}\n {{- tool_call.arguments | tojson }}\n {%- endif %}\n {{- '}\\n</tool_call>' }}\n {%- endfor %}\n {%- endif %}\n {{- '<|im_end|>\\n' }}\n {%- elif message.role == \"tool\" %}\n {%- if loop.first or (messages[loop.index0 - 1].role != \"tool\") %}\n {{- '<|im_start|>user' }}\n {%- endif %}\n {{- '\\n<tool_response>\\n' }}\n {{- content }}\n {{- '\\n</tool_response>' }}\n {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n {{- '<|im_end|>\\n' }}\n {%- endif %}\n {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n {{- '<|im_start|>assistant\n<think>\n' }}\n{%- endif %}",
6
+ "system_role_name": "system",
7
+ "user_role_name": "user",
8
+ "assistant_role_name": "assistant",
9
+ "eos_token": "<|im_end|>",
10
+ "bos_token": null,
11
+ "default_system_prompt": null,
12
+ "reasoning_config": {
13
+ "default_reasoning_effort": "medium",
14
+ "field_name": "enable_thinking",
15
+ "reasoning_effort_to_field_value": {
16
+ "medium": true,
17
+ "no_reasoning": false
18
+ }
19
+ },
20
+ "output_parser_regex": null,
21
+ "end_of_thinking_tag": null
22
+ },
23
+ "decoder_config": {
24
+ "embedding_config": {
25
+ "type": "UntiedEmbeddingConfig",
26
+ "input_scale": null,
27
+ "logit_soft_cap": null,
28
+ "logit_scale": null
29
+ },
30
+ "transformer_config": {
31
+ "layer_configs": [
32
+ {
33
+ "pre_mixer_norm_config": {
34
+ "epsilon": 1e-06,
35
+ "scale_offset": 1.0,
36
+ "upcast_mode": "only_normalization",
37
+ "subtract_mean": false,
38
+ "has_scale": true,
39
+ "has_biases": false
40
+ },
41
+ "mixer_conv_config": null,
42
+ "mixer_config": {
43
+ "type": "DeltaNetConfig",
44
+ "in_proj_config": {},
45
+ "conv_config": {
46
+ "has_biases": false
47
+ },
48
+ "out_proj_config": {},
49
+ "norm_config": {
50
+ "epsilon": 1e-06,
51
+ "scale_offset": null,
52
+ "upcast_mode": "only_normalization",
53
+ "subtract_mean": false,
54
+ "has_scale": true,
55
+ "has_biases": false
56
+ },
57
+ "num_heads": 32,
58
+ "num_groups": 16,
59
+ "head_dim": 128,
60
+ "value_head_dim": 128,
61
+ "kernel_size": 4
62
+ },
63
+ "post_mixer_norm_config": null,
64
+ "pre_mlp_norm_config": {
65
+ "epsilon": 1e-06,
66
+ "scale_offset": 1.0,
67
+ "upcast_mode": "only_normalization",
68
+ "subtract_mean": false,
69
+ "has_scale": true,
70
+ "has_biases": false
71
+ },
72
+ "mlp_conv_config": null,
73
+ "mlp_config": {
74
+ "type": "DenseMLPConfig",
75
+ "linear_config": {},
76
+ "activation": {
77
+ "type": "SiLU",
78
+ "alpha": 1.0
79
+ },
80
+ "has_up_biases": false,
81
+ "has_down_biases": false,
82
+ "gate_clipping": null,
83
+ "up_clipping": null
84
+ },
85
+ "post_mlp_norm_config": null,
86
+ "hidden_dim": null,
87
+ "ple_config": null,
88
+ "has_post_layer_scalar": false,
89
+ "kv_source_layer_index": null,
90
+ "rope_config": null
91
+ },
92
+ {
93
+ "pre_mixer_norm_config": {
94
+ "epsilon": 1e-06,
95
+ "scale_offset": 1.0,
96
+ "upcast_mode": "only_normalization",
97
+ "subtract_mean": false,
98
+ "has_scale": true,
99
+ "has_biases": false
100
+ },
101
+ "mixer_conv_config": null,
102
+ "mixer_config": {
103
+ "type": "DeltaNetConfig",
104
+ "in_proj_config": {},
105
+ "conv_config": {
106
+ "has_biases": false
107
+ },
108
+ "out_proj_config": {},
109
+ "norm_config": {
110
+ "epsilon": 1e-06,
111
+ "scale_offset": null,
112
+ "upcast_mode": "only_normalization",
113
+ "subtract_mean": false,
114
+ "has_scale": true,
115
+ "has_biases": false
116
+ },
117
+ "num_heads": 32,
118
+ "num_groups": 16,
119
+ "head_dim": 128,
120
+ "value_head_dim": 128,
121
+ "kernel_size": 4
122
+ },
123
+ "post_mixer_norm_config": null,
124
+ "pre_mlp_norm_config": {
125
+ "epsilon": 1e-06,
126
+ "scale_offset": 1.0,
127
+ "upcast_mode": "only_normalization",
128
+ "subtract_mean": false,
129
+ "has_scale": true,
130
+ "has_biases": false
131
+ },
132
+ "mlp_conv_config": null,
133
+ "mlp_config": {
134
+ "type": "DenseMLPConfig",
135
+ "linear_config": {},
136
+ "activation": {
137
+ "type": "SiLU",
138
+ "alpha": 1.0
139
+ },
140
+ "has_up_biases": false,
141
+ "has_down_biases": false,
142
+ "gate_clipping": null,
143
+ "up_clipping": null
144
+ },
145
+ "post_mlp_norm_config": null,
146
+ "hidden_dim": null,
147
+ "ple_config": null,
148
+ "has_post_layer_scalar": false,
149
+ "kv_source_layer_index": null,
150
+ "rope_config": null
151
+ },
152
+ {
153
+ "pre_mixer_norm_config": {
154
+ "epsilon": 1e-06,
155
+ "scale_offset": 1.0,
156
+ "upcast_mode": "only_normalization",
157
+ "subtract_mean": false,
158
+ "has_scale": true,
159
+ "has_biases": false
160
+ },
161
+ "mixer_conv_config": null,
162
+ "mixer_config": {
163
+ "type": "DeltaNetConfig",
164
+ "in_proj_config": {},
165
+ "conv_config": {
166
+ "has_biases": false
167
+ },
168
+ "out_proj_config": {},
169
+ "norm_config": {
170
+ "epsilon": 1e-06,
171
+ "scale_offset": null,
172
+ "upcast_mode": "only_normalization",
173
+ "subtract_mean": false,
174
+ "has_scale": true,
175
+ "has_biases": false
176
+ },
177
+ "num_heads": 32,
178
+ "num_groups": 16,
179
+ "head_dim": 128,
180
+ "value_head_dim": 128,
181
+ "kernel_size": 4
182
+ },
183
+ "post_mixer_norm_config": null,
184
+ "pre_mlp_norm_config": {
185
+ "epsilon": 1e-06,
186
+ "scale_offset": 1.0,
187
+ "upcast_mode": "only_normalization",
188
+ "subtract_mean": false,
189
+ "has_scale": true,
190
+ "has_biases": false
191
+ },
192
+ "mlp_conv_config": null,
193
+ "mlp_config": {
194
+ "type": "DenseMLPConfig",
195
+ "linear_config": {},
196
+ "activation": {
197
+ "type": "SiLU",
198
+ "alpha": 1.0
199
+ },
200
+ "has_up_biases": false,
201
+ "has_down_biases": false,
202
+ "gate_clipping": null,
203
+ "up_clipping": null
204
+ },
205
+ "post_mlp_norm_config": null,
206
+ "hidden_dim": null,
207
+ "ple_config": null,
208
+ "has_post_layer_scalar": false,
209
+ "kv_source_layer_index": null,
210
+ "rope_config": null
211
+ },
212
+ {
213
+ "pre_mixer_norm_config": {
214
+ "epsilon": 1e-06,
215
+ "scale_offset": 1.0,
216
+ "upcast_mode": "only_normalization",
217
+ "subtract_mean": false,
218
+ "has_scale": true,
219
+ "has_biases": false
220
+ },
221
+ "mixer_conv_config": null,
222
+ "mixer_config": {
223
+ "type": "AttentionConfig",
224
+ "qkvg_projection_config": {},
225
+ "out_projection_config": {},
226
+ "query_norm_config": {
227
+ "epsilon": 1e-06,
228
+ "scale_offset": 1.0,
229
+ "upcast_mode": "only_normalization",
230
+ "subtract_mean": false,
231
+ "has_scale": true,
232
+ "has_biases": false
233
+ },
234
+ "key_norm_config": {
235
+ "epsilon": 1e-06,
236
+ "scale_offset": 1.0,
237
+ "upcast_mode": "only_normalization",
238
+ "subtract_mean": false,
239
+ "has_scale": true,
240
+ "has_biases": false
241
+ },
242
+ "num_heads": 16,
243
+ "num_groups": 4,
244
+ "head_dim": 256,
245
+ "is_causal": true,
246
+ "scale": null,
247
+ "sliding_window_size": null,
248
+ "logit_soft_cap": null,
249
+ "has_sinks": false,
250
+ "has_qkvg_biases": false,
251
+ "has_out_biases": false,
252
+ "has_gate": true,
253
+ "normalize_values": false,
254
+ "is_kv_sharing": false
255
+ },
256
+ "post_mixer_norm_config": null,
257
+ "pre_mlp_norm_config": {
258
+ "epsilon": 1e-06,
259
+ "scale_offset": 1.0,
260
+ "upcast_mode": "only_normalization",
261
+ "subtract_mean": false,
262
+ "has_scale": true,
263
+ "has_biases": false
264
+ },
265
+ "mlp_conv_config": null,
266
+ "mlp_config": {
267
+ "type": "DenseMLPConfig",
268
+ "linear_config": {},
269
+ "activation": {
270
+ "type": "SiLU",
271
+ "alpha": 1.0
272
+ },
273
+ "has_up_biases": false,
274
+ "has_down_biases": false,
275
+ "gate_clipping": null,
276
+ "up_clipping": null
277
+ },
278
+ "post_mlp_norm_config": null,
279
+ "hidden_dim": null,
280
+ "ple_config": null,
281
+ "has_post_layer_scalar": false,
282
+ "kv_source_layer_index": null,
283
+ "rope_config": {
284
+ "type": "UnscaledRoPEConfig",
285
+ "base": 10000000.0,
286
+ "max_sequence_length": 262144,
287
+ "head_dim": 64
288
+ }
289
+ },
290
+ {
291
+ "pre_mixer_norm_config": {
292
+ "epsilon": 1e-06,
293
+ "scale_offset": 1.0,
294
+ "upcast_mode": "only_normalization",
295
+ "subtract_mean": false,
296
+ "has_scale": true,
297
+ "has_biases": false
298
+ },
299
+ "mixer_conv_config": null,
300
+ "mixer_config": {
301
+ "type": "DeltaNetConfig",
302
+ "in_proj_config": {},
303
+ "conv_config": {
304
+ "has_biases": false
305
+ },
306
+ "out_proj_config": {},
307
+ "norm_config": {
308
+ "epsilon": 1e-06,
309
+ "scale_offset": null,
310
+ "upcast_mode": "only_normalization",
311
+ "subtract_mean": false,
312
+ "has_scale": true,
313
+ "has_biases": false
314
+ },
315
+ "num_heads": 32,
316
+ "num_groups": 16,
317
+ "head_dim": 128,
318
+ "value_head_dim": 128,
319
+ "kernel_size": 4
320
+ },
321
+ "post_mixer_norm_config": null,
322
+ "pre_mlp_norm_config": {
323
+ "epsilon": 1e-06,
324
+ "scale_offset": 1.0,
325
+ "upcast_mode": "only_normalization",
326
+ "subtract_mean": false,
327
+ "has_scale": true,
328
+ "has_biases": false
329
+ },
330
+ "mlp_conv_config": null,
331
+ "mlp_config": {
332
+ "type": "DenseMLPConfig",
333
+ "linear_config": {},
334
+ "activation": {
335
+ "type": "SiLU",
336
+ "alpha": 1.0
337
+ },
338
+ "has_up_biases": false,
339
+ "has_down_biases": false,
340
+ "gate_clipping": null,
341
+ "up_clipping": null
342
+ },
343
+ "post_mlp_norm_config": null,
344
+ "hidden_dim": null,
345
+ "ple_config": null,
346
+ "has_post_layer_scalar": false,
347
+ "kv_source_layer_index": null,
348
+ "rope_config": null
349
+ },
350
+ {
351
+ "pre_mixer_norm_config": {
352
+ "epsilon": 1e-06,
353
+ "scale_offset": 1.0,
354
+ "upcast_mode": "only_normalization",
355
+ "subtract_mean": false,
356
+ "has_scale": true,
357
+ "has_biases": false
358
+ },
359
+ "mixer_conv_config": null,
360
+ "mixer_config": {
361
+ "type": "DeltaNetConfig",
362
+ "in_proj_config": {},
363
+ "conv_config": {
364
+ "has_biases": false
365
+ },
366
+ "out_proj_config": {},
367
+ "norm_config": {
368
+ "epsilon": 1e-06,
369
+ "scale_offset": null,
370
+ "upcast_mode": "only_normalization",
371
+ "subtract_mean": false,
372
+ "has_scale": true,
373
+ "has_biases": false
374
+ },
375
+ "num_heads": 32,
376
+ "num_groups": 16,
377
+ "head_dim": 128,
378
+ "value_head_dim": 128,
379
+ "kernel_size": 4
380
+ },
381
+ "post_mixer_norm_config": null,
382
+ "pre_mlp_norm_config": {
383
+ "epsilon": 1e-06,
384
+ "scale_offset": 1.0,
385
+ "upcast_mode": "only_normalization",
386
+ "subtract_mean": false,
387
+ "has_scale": true,
388
+ "has_biases": false
389
+ },
390
+ "mlp_conv_config": null,
391
+ "mlp_config": {
392
+ "type": "DenseMLPConfig",
393
+ "linear_config": {},
394
+ "activation": {
395
+ "type": "SiLU",
396
+ "alpha": 1.0
397
+ },
398
+ "has_up_biases": false,
399
+ "has_down_biases": false,
400
+ "gate_clipping": null,
401
+ "up_clipping": null
402
+ },
403
+ "post_mlp_norm_config": null,
404
+ "hidden_dim": null,
405
+ "ple_config": null,
406
+ "has_post_layer_scalar": false,
407
+ "kv_source_layer_index": null,
408
+ "rope_config": null
409
+ },
410
+ {
411
+ "pre_mixer_norm_config": {
412
+ "epsilon": 1e-06,
413
+ "scale_offset": 1.0,
414
+ "upcast_mode": "only_normalization",
415
+ "subtract_mean": false,
416
+ "has_scale": true,
417
+ "has_biases": false
418
+ },
419
+ "mixer_conv_config": null,
420
+ "mixer_config": {
421
+ "type": "DeltaNetConfig",
422
+ "in_proj_config": {},
423
+ "conv_config": {
424
+ "has_biases": false
425
+ },
426
+ "out_proj_config": {},
427
+ "norm_config": {
428
+ "epsilon": 1e-06,
429
+ "scale_offset": null,
430
+ "upcast_mode": "only_normalization",
431
+ "subtract_mean": false,
432
+ "has_scale": true,
433
+ "has_biases": false
434
+ },
435
+ "num_heads": 32,
436
+ "num_groups": 16,
437
+ "head_dim": 128,
438
+ "value_head_dim": 128,
439
+ "kernel_size": 4
440
+ },
441
+ "post_mixer_norm_config": null,
442
+ "pre_mlp_norm_config": {
443
+ "epsilon": 1e-06,
444
+ "scale_offset": 1.0,
445
+ "upcast_mode": "only_normalization",
446
+ "subtract_mean": false,
447
+ "has_scale": true,
448
+ "has_biases": false
449
+ },
450
+ "mlp_conv_config": null,
451
+ "mlp_config": {
452
+ "type": "DenseMLPConfig",
453
+ "linear_config": {},
454
+ "activation": {
455
+ "type": "SiLU",
456
+ "alpha": 1.0
457
+ },
458
+ "has_up_biases": false,
459
+ "has_down_biases": false,
460
+ "gate_clipping": null,
461
+ "up_clipping": null
462
+ },
463
+ "post_mlp_norm_config": null,
464
+ "hidden_dim": null,
465
+ "ple_config": null,
466
+ "has_post_layer_scalar": false,
467
+ "kv_source_layer_index": null,
468
+ "rope_config": null
469
+ },
470
+ {
471
+ "pre_mixer_norm_config": {
472
+ "epsilon": 1e-06,
473
+ "scale_offset": 1.0,
474
+ "upcast_mode": "only_normalization",
475
+ "subtract_mean": false,
476
+ "has_scale": true,
477
+ "has_biases": false
478
+ },
479
+ "mixer_conv_config": null,
480
+ "mixer_config": {
481
+ "type": "AttentionConfig",
482
+ "qkvg_projection_config": {},
483
+ "out_projection_config": {},
484
+ "query_norm_config": {
485
+ "epsilon": 1e-06,
486
+ "scale_offset": 1.0,
487
+ "upcast_mode": "only_normalization",
488
+ "subtract_mean": false,
489
+ "has_scale": true,
490
+ "has_biases": false
491
+ },
492
+ "key_norm_config": {
493
+ "epsilon": 1e-06,
494
+ "scale_offset": 1.0,
495
+ "upcast_mode": "only_normalization",
496
+ "subtract_mean": false,
497
+ "has_scale": true,
498
+ "has_biases": false
499
+ },
500
+ "num_heads": 16,
501
+ "num_groups": 4,
502
+ "head_dim": 256,
503
+ "is_causal": true,
504
+ "scale": null,
505
+ "sliding_window_size": null,
506
+ "logit_soft_cap": null,
507
+ "has_sinks": false,
508
+ "has_qkvg_biases": false,
509
+ "has_out_biases": false,
510
+ "has_gate": true,
511
+ "normalize_values": false,
512
+ "is_kv_sharing": false
513
+ },
514
+ "post_mixer_norm_config": null,
515
+ "pre_mlp_norm_config": {
516
+ "epsilon": 1e-06,
517
+ "scale_offset": 1.0,
518
+ "upcast_mode": "only_normalization",
519
+ "subtract_mean": false,
520
+ "has_scale": true,
521
+ "has_biases": false
522
+ },
523
+ "mlp_conv_config": null,
524
+ "mlp_config": {
525
+ "type": "DenseMLPConfig",
526
+ "linear_config": {},
527
+ "activation": {
528
+ "type": "SiLU",
529
+ "alpha": 1.0
530
+ },
531
+ "has_up_biases": false,
532
+ "has_down_biases": false,
533
+ "gate_clipping": null,
534
+ "up_clipping": null
535
+ },
536
+ "post_mlp_norm_config": null,
537
+ "hidden_dim": null,
538
+ "ple_config": null,
539
+ "has_post_layer_scalar": false,
540
+ "kv_source_layer_index": null,
541
+ "rope_config": {
542
+ "type": "UnscaledRoPEConfig",
543
+ "base": 10000000.0,
544
+ "max_sequence_length": 262144,
545
+ "head_dim": 64
546
+ }
547
+ },
548
+ {
549
+ "pre_mixer_norm_config": {
550
+ "epsilon": 1e-06,
551
+ "scale_offset": 1.0,
552
+ "upcast_mode": "only_normalization",
553
+ "subtract_mean": false,
554
+ "has_scale": true,
555
+ "has_biases": false
556
+ },
557
+ "mixer_conv_config": null,
558
+ "mixer_config": {
559
+ "type": "DeltaNetConfig",
560
+ "in_proj_config": {},
561
+ "conv_config": {
562
+ "has_biases": false
563
+ },
564
+ "out_proj_config": {},
565
+ "norm_config": {
566
+ "epsilon": 1e-06,
567
+ "scale_offset": null,
568
+ "upcast_mode": "only_normalization",
569
+ "subtract_mean": false,
570
+ "has_scale": true,
571
+ "has_biases": false
572
+ },
573
+ "num_heads": 32,
574
+ "num_groups": 16,
575
+ "head_dim": 128,
576
+ "value_head_dim": 128,
577
+ "kernel_size": 4
578
+ },
579
+ "post_mixer_norm_config": null,
580
+ "pre_mlp_norm_config": {
581
+ "epsilon": 1e-06,
582
+ "scale_offset": 1.0,
583
+ "upcast_mode": "only_normalization",
584
+ "subtract_mean": false,
585
+ "has_scale": true,
586
+ "has_biases": false
587
+ },
588
+ "mlp_conv_config": null,
589
+ "mlp_config": {
590
+ "type": "DenseMLPConfig",
591
+ "linear_config": {},
592
+ "activation": {
593
+ "type": "SiLU",
594
+ "alpha": 1.0
595
+ },
596
+ "has_up_biases": false,
597
+ "has_down_biases": false,
598
+ "gate_clipping": null,
599
+ "up_clipping": null
600
+ },
601
+ "post_mlp_norm_config": null,
602
+ "hidden_dim": null,
603
+ "ple_config": null,
604
+ "has_post_layer_scalar": false,
605
+ "kv_source_layer_index": null,
606
+ "rope_config": null
607
+ },
608
+ {
609
+ "pre_mixer_norm_config": {
610
+ "epsilon": 1e-06,
611
+ "scale_offset": 1.0,
612
+ "upcast_mode": "only_normalization",
613
+ "subtract_mean": false,
614
+ "has_scale": true,
615
+ "has_biases": false
616
+ },
617
+ "mixer_conv_config": null,
618
+ "mixer_config": {
619
+ "type": "DeltaNetConfig",
620
+ "in_proj_config": {},
621
+ "conv_config": {
622
+ "has_biases": false
623
+ },
624
+ "out_proj_config": {},
625
+ "norm_config": {
626
+ "epsilon": 1e-06,
627
+ "scale_offset": null,
628
+ "upcast_mode": "only_normalization",
629
+ "subtract_mean": false,
630
+ "has_scale": true,
631
+ "has_biases": false
632
+ },
633
+ "num_heads": 32,
634
+ "num_groups": 16,
635
+ "head_dim": 128,
636
+ "value_head_dim": 128,
637
+ "kernel_size": 4
638
+ },
639
+ "post_mixer_norm_config": null,
640
+ "pre_mlp_norm_config": {
641
+ "epsilon": 1e-06,
642
+ "scale_offset": 1.0,
643
+ "upcast_mode": "only_normalization",
644
+ "subtract_mean": false,
645
+ "has_scale": true,
646
+ "has_biases": false
647
+ },
648
+ "mlp_conv_config": null,
649
+ "mlp_config": {
650
+ "type": "DenseMLPConfig",
651
+ "linear_config": {},
652
+ "activation": {
653
+ "type": "SiLU",
654
+ "alpha": 1.0
655
+ },
656
+ "has_up_biases": false,
657
+ "has_down_biases": false,
658
+ "gate_clipping": null,
659
+ "up_clipping": null
660
+ },
661
+ "post_mlp_norm_config": null,
662
+ "hidden_dim": null,
663
+ "ple_config": null,
664
+ "has_post_layer_scalar": false,
665
+ "kv_source_layer_index": null,
666
+ "rope_config": null
667
+ },
668
+ {
669
+ "pre_mixer_norm_config": {
670
+ "epsilon": 1e-06,
671
+ "scale_offset": 1.0,
672
+ "upcast_mode": "only_normalization",
673
+ "subtract_mean": false,
674
+ "has_scale": true,
675
+ "has_biases": false
676
+ },
677
+ "mixer_conv_config": null,
678
+ "mixer_config": {
679
+ "type": "DeltaNetConfig",
680
+ "in_proj_config": {},
681
+ "conv_config": {
682
+ "has_biases": false
683
+ },
684
+ "out_proj_config": {},
685
+ "norm_config": {
686
+ "epsilon": 1e-06,
687
+ "scale_offset": null,
688
+ "upcast_mode": "only_normalization",
689
+ "subtract_mean": false,
690
+ "has_scale": true,
691
+ "has_biases": false
692
+ },
693
+ "num_heads": 32,
694
+ "num_groups": 16,
695
+ "head_dim": 128,
696
+ "value_head_dim": 128,
697
+ "kernel_size": 4
698
+ },
699
+ "post_mixer_norm_config": null,
700
+ "pre_mlp_norm_config": {
701
+ "epsilon": 1e-06,
702
+ "scale_offset": 1.0,
703
+ "upcast_mode": "only_normalization",
704
+ "subtract_mean": false,
705
+ "has_scale": true,
706
+ "has_biases": false
707
+ },
708
+ "mlp_conv_config": null,
709
+ "mlp_config": {
710
+ "type": "DenseMLPConfig",
711
+ "linear_config": {},
712
+ "activation": {
713
+ "type": "SiLU",
714
+ "alpha": 1.0
715
+ },
716
+ "has_up_biases": false,
717
+ "has_down_biases": false,
718
+ "gate_clipping": null,
719
+ "up_clipping": null
720
+ },
721
+ "post_mlp_norm_config": null,
722
+ "hidden_dim": null,
723
+ "ple_config": null,
724
+ "has_post_layer_scalar": false,
725
+ "kv_source_layer_index": null,
726
+ "rope_config": null
727
+ },
728
+ {
729
+ "pre_mixer_norm_config": {
730
+ "epsilon": 1e-06,
731
+ "scale_offset": 1.0,
732
+ "upcast_mode": "only_normalization",
733
+ "subtract_mean": false,
734
+ "has_scale": true,
735
+ "has_biases": false
736
+ },
737
+ "mixer_conv_config": null,
738
+ "mixer_config": {
739
+ "type": "AttentionConfig",
740
+ "qkvg_projection_config": {},
741
+ "out_projection_config": {},
742
+ "query_norm_config": {
743
+ "epsilon": 1e-06,
744
+ "scale_offset": 1.0,
745
+ "upcast_mode": "only_normalization",
746
+ "subtract_mean": false,
747
+ "has_scale": true,
748
+ "has_biases": false
749
+ },
750
+ "key_norm_config": {
751
+ "epsilon": 1e-06,
752
+ "scale_offset": 1.0,
753
+ "upcast_mode": "only_normalization",
754
+ "subtract_mean": false,
755
+ "has_scale": true,
756
+ "has_biases": false
757
+ },
758
+ "num_heads": 16,
759
+ "num_groups": 4,
760
+ "head_dim": 256,
761
+ "is_causal": true,
762
+ "scale": null,
763
+ "sliding_window_size": null,
764
+ "logit_soft_cap": null,
765
+ "has_sinks": false,
766
+ "has_qkvg_biases": false,
767
+ "has_out_biases": false,
768
+ "has_gate": true,
769
+ "normalize_values": false,
770
+ "is_kv_sharing": false
771
+ },
772
+ "post_mixer_norm_config": null,
773
+ "pre_mlp_norm_config": {
774
+ "epsilon": 1e-06,
775
+ "scale_offset": 1.0,
776
+ "upcast_mode": "only_normalization",
777
+ "subtract_mean": false,
778
+ "has_scale": true,
779
+ "has_biases": false
780
+ },
781
+ "mlp_conv_config": null,
782
+ "mlp_config": {
783
+ "type": "DenseMLPConfig",
784
+ "linear_config": {},
785
+ "activation": {
786
+ "type": "SiLU",
787
+ "alpha": 1.0
788
+ },
789
+ "has_up_biases": false,
790
+ "has_down_biases": false,
791
+ "gate_clipping": null,
792
+ "up_clipping": null
793
+ },
794
+ "post_mlp_norm_config": null,
795
+ "hidden_dim": null,
796
+ "ple_config": null,
797
+ "has_post_layer_scalar": false,
798
+ "kv_source_layer_index": null,
799
+ "rope_config": {
800
+ "type": "UnscaledRoPEConfig",
801
+ "base": 10000000.0,
802
+ "max_sequence_length": 262144,
803
+ "head_dim": 64
804
+ }
805
+ },
806
+ {
807
+ "pre_mixer_norm_config": {
808
+ "epsilon": 1e-06,
809
+ "scale_offset": 1.0,
810
+ "upcast_mode": "only_normalization",
811
+ "subtract_mean": false,
812
+ "has_scale": true,
813
+ "has_biases": false
814
+ },
815
+ "mixer_conv_config": null,
816
+ "mixer_config": {
817
+ "type": "DeltaNetConfig",
818
+ "in_proj_config": {},
819
+ "conv_config": {
820
+ "has_biases": false
821
+ },
822
+ "out_proj_config": {},
823
+ "norm_config": {
824
+ "epsilon": 1e-06,
825
+ "scale_offset": null,
826
+ "upcast_mode": "only_normalization",
827
+ "subtract_mean": false,
828
+ "has_scale": true,
829
+ "has_biases": false
830
+ },
831
+ "num_heads": 32,
832
+ "num_groups": 16,
833
+ "head_dim": 128,
834
+ "value_head_dim": 128,
835
+ "kernel_size": 4
836
+ },
837
+ "post_mixer_norm_config": null,
838
+ "pre_mlp_norm_config": {
839
+ "epsilon": 1e-06,
840
+ "scale_offset": 1.0,
841
+ "upcast_mode": "only_normalization",
842
+ "subtract_mean": false,
843
+ "has_scale": true,
844
+ "has_biases": false
845
+ },
846
+ "mlp_conv_config": null,
847
+ "mlp_config": {
848
+ "type": "DenseMLPConfig",
849
+ "linear_config": {},
850
+ "activation": {
851
+ "type": "SiLU",
852
+ "alpha": 1.0
853
+ },
854
+ "has_up_biases": false,
855
+ "has_down_biases": false,
856
+ "gate_clipping": null,
857
+ "up_clipping": null
858
+ },
859
+ "post_mlp_norm_config": null,
860
+ "hidden_dim": null,
861
+ "ple_config": null,
862
+ "has_post_layer_scalar": false,
863
+ "kv_source_layer_index": null,
864
+ "rope_config": null
865
+ },
866
+ {
867
+ "pre_mixer_norm_config": {
868
+ "epsilon": 1e-06,
869
+ "scale_offset": 1.0,
870
+ "upcast_mode": "only_normalization",
871
+ "subtract_mean": false,
872
+ "has_scale": true,
873
+ "has_biases": false
874
+ },
875
+ "mixer_conv_config": null,
876
+ "mixer_config": {
877
+ "type": "DeltaNetConfig",
878
+ "in_proj_config": {},
879
+ "conv_config": {
880
+ "has_biases": false
881
+ },
882
+ "out_proj_config": {},
883
+ "norm_config": {
884
+ "epsilon": 1e-06,
885
+ "scale_offset": null,
886
+ "upcast_mode": "only_normalization",
887
+ "subtract_mean": false,
888
+ "has_scale": true,
889
+ "has_biases": false
890
+ },
891
+ "num_heads": 32,
892
+ "num_groups": 16,
893
+ "head_dim": 128,
894
+ "value_head_dim": 128,
895
+ "kernel_size": 4
896
+ },
897
+ "post_mixer_norm_config": null,
898
+ "pre_mlp_norm_config": {
899
+ "epsilon": 1e-06,
900
+ "scale_offset": 1.0,
901
+ "upcast_mode": "only_normalization",
902
+ "subtract_mean": false,
903
+ "has_scale": true,
904
+ "has_biases": false
905
+ },
906
+ "mlp_conv_config": null,
907
+ "mlp_config": {
908
+ "type": "DenseMLPConfig",
909
+ "linear_config": {},
910
+ "activation": {
911
+ "type": "SiLU",
912
+ "alpha": 1.0
913
+ },
914
+ "has_up_biases": false,
915
+ "has_down_biases": false,
916
+ "gate_clipping": null,
917
+ "up_clipping": null
918
+ },
919
+ "post_mlp_norm_config": null,
920
+ "hidden_dim": null,
921
+ "ple_config": null,
922
+ "has_post_layer_scalar": false,
923
+ "kv_source_layer_index": null,
924
+ "rope_config": null
925
+ },
926
+ {
927
+ "pre_mixer_norm_config": {
928
+ "epsilon": 1e-06,
929
+ "scale_offset": 1.0,
930
+ "upcast_mode": "only_normalization",
931
+ "subtract_mean": false,
932
+ "has_scale": true,
933
+ "has_biases": false
934
+ },
935
+ "mixer_conv_config": null,
936
+ "mixer_config": {
937
+ "type": "DeltaNetConfig",
938
+ "in_proj_config": {},
939
+ "conv_config": {
940
+ "has_biases": false
941
+ },
942
+ "out_proj_config": {},
943
+ "norm_config": {
944
+ "epsilon": 1e-06,
945
+ "scale_offset": null,
946
+ "upcast_mode": "only_normalization",
947
+ "subtract_mean": false,
948
+ "has_scale": true,
949
+ "has_biases": false
950
+ },
951
+ "num_heads": 32,
952
+ "num_groups": 16,
953
+ "head_dim": 128,
954
+ "value_head_dim": 128,
955
+ "kernel_size": 4
956
+ },
957
+ "post_mixer_norm_config": null,
958
+ "pre_mlp_norm_config": {
959
+ "epsilon": 1e-06,
960
+ "scale_offset": 1.0,
961
+ "upcast_mode": "only_normalization",
962
+ "subtract_mean": false,
963
+ "has_scale": true,
964
+ "has_biases": false
965
+ },
966
+ "mlp_conv_config": null,
967
+ "mlp_config": {
968
+ "type": "DenseMLPConfig",
969
+ "linear_config": {},
970
+ "activation": {
971
+ "type": "SiLU",
972
+ "alpha": 1.0
973
+ },
974
+ "has_up_biases": false,
975
+ "has_down_biases": false,
976
+ "gate_clipping": null,
977
+ "up_clipping": null
978
+ },
979
+ "post_mlp_norm_config": null,
980
+ "hidden_dim": null,
981
+ "ple_config": null,
982
+ "has_post_layer_scalar": false,
983
+ "kv_source_layer_index": null,
984
+ "rope_config": null
985
+ },
986
+ {
987
+ "pre_mixer_norm_config": {
988
+ "epsilon": 1e-06,
989
+ "scale_offset": 1.0,
990
+ "upcast_mode": "only_normalization",
991
+ "subtract_mean": false,
992
+ "has_scale": true,
993
+ "has_biases": false
994
+ },
995
+ "mixer_conv_config": null,
996
+ "mixer_config": {
997
+ "type": "AttentionConfig",
998
+ "qkvg_projection_config": {},
999
+ "out_projection_config": {},
1000
+ "query_norm_config": {
1001
+ "epsilon": 1e-06,
1002
+ "scale_offset": 1.0,
1003
+ "upcast_mode": "only_normalization",
1004
+ "subtract_mean": false,
1005
+ "has_scale": true,
1006
+ "has_biases": false
1007
+ },
1008
+ "key_norm_config": {
1009
+ "epsilon": 1e-06,
1010
+ "scale_offset": 1.0,
1011
+ "upcast_mode": "only_normalization",
1012
+ "subtract_mean": false,
1013
+ "has_scale": true,
1014
+ "has_biases": false
1015
+ },
1016
+ "num_heads": 16,
1017
+ "num_groups": 4,
1018
+ "head_dim": 256,
1019
+ "is_causal": true,
1020
+ "scale": null,
1021
+ "sliding_window_size": null,
1022
+ "logit_soft_cap": null,
1023
+ "has_sinks": false,
1024
+ "has_qkvg_biases": false,
1025
+ "has_out_biases": false,
1026
+ "has_gate": true,
1027
+ "normalize_values": false,
1028
+ "is_kv_sharing": false
1029
+ },
1030
+ "post_mixer_norm_config": null,
1031
+ "pre_mlp_norm_config": {
1032
+ "epsilon": 1e-06,
1033
+ "scale_offset": 1.0,
1034
+ "upcast_mode": "only_normalization",
1035
+ "subtract_mean": false,
1036
+ "has_scale": true,
1037
+ "has_biases": false
1038
+ },
1039
+ "mlp_conv_config": null,
1040
+ "mlp_config": {
1041
+ "type": "DenseMLPConfig",
1042
+ "linear_config": {},
1043
+ "activation": {
1044
+ "type": "SiLU",
1045
+ "alpha": 1.0
1046
+ },
1047
+ "has_up_biases": false,
1048
+ "has_down_biases": false,
1049
+ "gate_clipping": null,
1050
+ "up_clipping": null
1051
+ },
1052
+ "post_mlp_norm_config": null,
1053
+ "hidden_dim": null,
1054
+ "ple_config": null,
1055
+ "has_post_layer_scalar": false,
1056
+ "kv_source_layer_index": null,
1057
+ "rope_config": {
1058
+ "type": "UnscaledRoPEConfig",
1059
+ "base": 10000000.0,
1060
+ "max_sequence_length": 262144,
1061
+ "head_dim": 64
1062
+ }
1063
+ },
1064
+ {
1065
+ "pre_mixer_norm_config": {
1066
+ "epsilon": 1e-06,
1067
+ "scale_offset": 1.0,
1068
+ "upcast_mode": "only_normalization",
1069
+ "subtract_mean": false,
1070
+ "has_scale": true,
1071
+ "has_biases": false
1072
+ },
1073
+ "mixer_conv_config": null,
1074
+ "mixer_config": {
1075
+ "type": "DeltaNetConfig",
1076
+ "in_proj_config": {},
1077
+ "conv_config": {
1078
+ "has_biases": false
1079
+ },
1080
+ "out_proj_config": {},
1081
+ "norm_config": {
1082
+ "epsilon": 1e-06,
1083
+ "scale_offset": null,
1084
+ "upcast_mode": "only_normalization",
1085
+ "subtract_mean": false,
1086
+ "has_scale": true,
1087
+ "has_biases": false
1088
+ },
1089
+ "num_heads": 32,
1090
+ "num_groups": 16,
1091
+ "head_dim": 128,
1092
+ "value_head_dim": 128,
1093
+ "kernel_size": 4
1094
+ },
1095
+ "post_mixer_norm_config": null,
1096
+ "pre_mlp_norm_config": {
1097
+ "epsilon": 1e-06,
1098
+ "scale_offset": 1.0,
1099
+ "upcast_mode": "only_normalization",
1100
+ "subtract_mean": false,
1101
+ "has_scale": true,
1102
+ "has_biases": false
1103
+ },
1104
+ "mlp_conv_config": null,
1105
+ "mlp_config": {
1106
+ "type": "DenseMLPConfig",
1107
+ "linear_config": {},
1108
+ "activation": {
1109
+ "type": "SiLU",
1110
+ "alpha": 1.0
1111
+ },
1112
+ "has_up_biases": false,
1113
+ "has_down_biases": false,
1114
+ "gate_clipping": null,
1115
+ "up_clipping": null
1116
+ },
1117
+ "post_mlp_norm_config": null,
1118
+ "hidden_dim": null,
1119
+ "ple_config": null,
1120
+ "has_post_layer_scalar": false,
1121
+ "kv_source_layer_index": null,
1122
+ "rope_config": null
1123
+ },
1124
+ {
1125
+ "pre_mixer_norm_config": {
1126
+ "epsilon": 1e-06,
1127
+ "scale_offset": 1.0,
1128
+ "upcast_mode": "only_normalization",
1129
+ "subtract_mean": false,
1130
+ "has_scale": true,
1131
+ "has_biases": false
1132
+ },
1133
+ "mixer_conv_config": null,
1134
+ "mixer_config": {
1135
+ "type": "DeltaNetConfig",
1136
+ "in_proj_config": {},
1137
+ "conv_config": {
1138
+ "has_biases": false
1139
+ },
1140
+ "out_proj_config": {},
1141
+ "norm_config": {
1142
+ "epsilon": 1e-06,
1143
+ "scale_offset": null,
1144
+ "upcast_mode": "only_normalization",
1145
+ "subtract_mean": false,
1146
+ "has_scale": true,
1147
+ "has_biases": false
1148
+ },
1149
+ "num_heads": 32,
1150
+ "num_groups": 16,
1151
+ "head_dim": 128,
1152
+ "value_head_dim": 128,
1153
+ "kernel_size": 4
1154
+ },
1155
+ "post_mixer_norm_config": null,
1156
+ "pre_mlp_norm_config": {
1157
+ "epsilon": 1e-06,
1158
+ "scale_offset": 1.0,
1159
+ "upcast_mode": "only_normalization",
1160
+ "subtract_mean": false,
1161
+ "has_scale": true,
1162
+ "has_biases": false
1163
+ },
1164
+ "mlp_conv_config": null,
1165
+ "mlp_config": {
1166
+ "type": "DenseMLPConfig",
1167
+ "linear_config": {},
1168
+ "activation": {
1169
+ "type": "SiLU",
1170
+ "alpha": 1.0
1171
+ },
1172
+ "has_up_biases": false,
1173
+ "has_down_biases": false,
1174
+ "gate_clipping": null,
1175
+ "up_clipping": null
1176
+ },
1177
+ "post_mlp_norm_config": null,
1178
+ "hidden_dim": null,
1179
+ "ple_config": null,
1180
+ "has_post_layer_scalar": false,
1181
+ "kv_source_layer_index": null,
1182
+ "rope_config": null
1183
+ },
1184
+ {
1185
+ "pre_mixer_norm_config": {
1186
+ "epsilon": 1e-06,
1187
+ "scale_offset": 1.0,
1188
+ "upcast_mode": "only_normalization",
1189
+ "subtract_mean": false,
1190
+ "has_scale": true,
1191
+ "has_biases": false
1192
+ },
1193
+ "mixer_conv_config": null,
1194
+ "mixer_config": {
1195
+ "type": "DeltaNetConfig",
1196
+ "in_proj_config": {},
1197
+ "conv_config": {
1198
+ "has_biases": false
1199
+ },
1200
+ "out_proj_config": {},
1201
+ "norm_config": {
1202
+ "epsilon": 1e-06,
1203
+ "scale_offset": null,
1204
+ "upcast_mode": "only_normalization",
1205
+ "subtract_mean": false,
1206
+ "has_scale": true,
1207
+ "has_biases": false
1208
+ },
1209
+ "num_heads": 32,
1210
+ "num_groups": 16,
1211
+ "head_dim": 128,
1212
+ "value_head_dim": 128,
1213
+ "kernel_size": 4
1214
+ },
1215
+ "post_mixer_norm_config": null,
1216
+ "pre_mlp_norm_config": {
1217
+ "epsilon": 1e-06,
1218
+ "scale_offset": 1.0,
1219
+ "upcast_mode": "only_normalization",
1220
+ "subtract_mean": false,
1221
+ "has_scale": true,
1222
+ "has_biases": false
1223
+ },
1224
+ "mlp_conv_config": null,
1225
+ "mlp_config": {
1226
+ "type": "DenseMLPConfig",
1227
+ "linear_config": {},
1228
+ "activation": {
1229
+ "type": "SiLU",
1230
+ "alpha": 1.0
1231
+ },
1232
+ "has_up_biases": false,
1233
+ "has_down_biases": false,
1234
+ "gate_clipping": null,
1235
+ "up_clipping": null
1236
+ },
1237
+ "post_mlp_norm_config": null,
1238
+ "hidden_dim": null,
1239
+ "ple_config": null,
1240
+ "has_post_layer_scalar": false,
1241
+ "kv_source_layer_index": null,
1242
+ "rope_config": null
1243
+ },
1244
+ {
1245
+ "pre_mixer_norm_config": {
1246
+ "epsilon": 1e-06,
1247
+ "scale_offset": 1.0,
1248
+ "upcast_mode": "only_normalization",
1249
+ "subtract_mean": false,
1250
+ "has_scale": true,
1251
+ "has_biases": false
1252
+ },
1253
+ "mixer_conv_config": null,
1254
+ "mixer_config": {
1255
+ "type": "AttentionConfig",
1256
+ "qkvg_projection_config": {},
1257
+ "out_projection_config": {},
1258
+ "query_norm_config": {
1259
+ "epsilon": 1e-06,
1260
+ "scale_offset": 1.0,
1261
+ "upcast_mode": "only_normalization",
1262
+ "subtract_mean": false,
1263
+ "has_scale": true,
1264
+ "has_biases": false
1265
+ },
1266
+ "key_norm_config": {
1267
+ "epsilon": 1e-06,
1268
+ "scale_offset": 1.0,
1269
+ "upcast_mode": "only_normalization",
1270
+ "subtract_mean": false,
1271
+ "has_scale": true,
1272
+ "has_biases": false
1273
+ },
1274
+ "num_heads": 16,
1275
+ "num_groups": 4,
1276
+ "head_dim": 256,
1277
+ "is_causal": true,
1278
+ "scale": null,
1279
+ "sliding_window_size": null,
1280
+ "logit_soft_cap": null,
1281
+ "has_sinks": false,
1282
+ "has_qkvg_biases": false,
1283
+ "has_out_biases": false,
1284
+ "has_gate": true,
1285
+ "normalize_values": false,
1286
+ "is_kv_sharing": false
1287
+ },
1288
+ "post_mixer_norm_config": null,
1289
+ "pre_mlp_norm_config": {
1290
+ "epsilon": 1e-06,
1291
+ "scale_offset": 1.0,
1292
+ "upcast_mode": "only_normalization",
1293
+ "subtract_mean": false,
1294
+ "has_scale": true,
1295
+ "has_biases": false
1296
+ },
1297
+ "mlp_conv_config": null,
1298
+ "mlp_config": {
1299
+ "type": "DenseMLPConfig",
1300
+ "linear_config": {},
1301
+ "activation": {
1302
+ "type": "SiLU",
1303
+ "alpha": 1.0
1304
+ },
1305
+ "has_up_biases": false,
1306
+ "has_down_biases": false,
1307
+ "gate_clipping": null,
1308
+ "up_clipping": null
1309
+ },
1310
+ "post_mlp_norm_config": null,
1311
+ "hidden_dim": null,
1312
+ "ple_config": null,
1313
+ "has_post_layer_scalar": false,
1314
+ "kv_source_layer_index": null,
1315
+ "rope_config": {
1316
+ "type": "UnscaledRoPEConfig",
1317
+ "base": 10000000.0,
1318
+ "max_sequence_length": 262144,
1319
+ "head_dim": 64
1320
+ }
1321
+ },
1322
+ {
1323
+ "pre_mixer_norm_config": {
1324
+ "epsilon": 1e-06,
1325
+ "scale_offset": 1.0,
1326
+ "upcast_mode": "only_normalization",
1327
+ "subtract_mean": false,
1328
+ "has_scale": true,
1329
+ "has_biases": false
1330
+ },
1331
+ "mixer_conv_config": null,
1332
+ "mixer_config": {
1333
+ "type": "DeltaNetConfig",
1334
+ "in_proj_config": {},
1335
+ "conv_config": {
1336
+ "has_biases": false
1337
+ },
1338
+ "out_proj_config": {},
1339
+ "norm_config": {
1340
+ "epsilon": 1e-06,
1341
+ "scale_offset": null,
1342
+ "upcast_mode": "only_normalization",
1343
+ "subtract_mean": false,
1344
+ "has_scale": true,
1345
+ "has_biases": false
1346
+ },
1347
+ "num_heads": 32,
1348
+ "num_groups": 16,
1349
+ "head_dim": 128,
1350
+ "value_head_dim": 128,
1351
+ "kernel_size": 4
1352
+ },
1353
+ "post_mixer_norm_config": null,
1354
+ "pre_mlp_norm_config": {
1355
+ "epsilon": 1e-06,
1356
+ "scale_offset": 1.0,
1357
+ "upcast_mode": "only_normalization",
1358
+ "subtract_mean": false,
1359
+ "has_scale": true,
1360
+ "has_biases": false
1361
+ },
1362
+ "mlp_conv_config": null,
1363
+ "mlp_config": {
1364
+ "type": "DenseMLPConfig",
1365
+ "linear_config": {},
1366
+ "activation": {
1367
+ "type": "SiLU",
1368
+ "alpha": 1.0
1369
+ },
1370
+ "has_up_biases": false,
1371
+ "has_down_biases": false,
1372
+ "gate_clipping": null,
1373
+ "up_clipping": null
1374
+ },
1375
+ "post_mlp_norm_config": null,
1376
+ "hidden_dim": null,
1377
+ "ple_config": null,
1378
+ "has_post_layer_scalar": false,
1379
+ "kv_source_layer_index": null,
1380
+ "rope_config": null
1381
+ },
1382
+ {
1383
+ "pre_mixer_norm_config": {
1384
+ "epsilon": 1e-06,
1385
+ "scale_offset": 1.0,
1386
+ "upcast_mode": "only_normalization",
1387
+ "subtract_mean": false,
1388
+ "has_scale": true,
1389
+ "has_biases": false
1390
+ },
1391
+ "mixer_conv_config": null,
1392
+ "mixer_config": {
1393
+ "type": "DeltaNetConfig",
1394
+ "in_proj_config": {},
1395
+ "conv_config": {
1396
+ "has_biases": false
1397
+ },
1398
+ "out_proj_config": {},
1399
+ "norm_config": {
1400
+ "epsilon": 1e-06,
1401
+ "scale_offset": null,
1402
+ "upcast_mode": "only_normalization",
1403
+ "subtract_mean": false,
1404
+ "has_scale": true,
1405
+ "has_biases": false
1406
+ },
1407
+ "num_heads": 32,
1408
+ "num_groups": 16,
1409
+ "head_dim": 128,
1410
+ "value_head_dim": 128,
1411
+ "kernel_size": 4
1412
+ },
1413
+ "post_mixer_norm_config": null,
1414
+ "pre_mlp_norm_config": {
1415
+ "epsilon": 1e-06,
1416
+ "scale_offset": 1.0,
1417
+ "upcast_mode": "only_normalization",
1418
+ "subtract_mean": false,
1419
+ "has_scale": true,
1420
+ "has_biases": false
1421
+ },
1422
+ "mlp_conv_config": null,
1423
+ "mlp_config": {
1424
+ "type": "DenseMLPConfig",
1425
+ "linear_config": {},
1426
+ "activation": {
1427
+ "type": "SiLU",
1428
+ "alpha": 1.0
1429
+ },
1430
+ "has_up_biases": false,
1431
+ "has_down_biases": false,
1432
+ "gate_clipping": null,
1433
+ "up_clipping": null
1434
+ },
1435
+ "post_mlp_norm_config": null,
1436
+ "hidden_dim": null,
1437
+ "ple_config": null,
1438
+ "has_post_layer_scalar": false,
1439
+ "kv_source_layer_index": null,
1440
+ "rope_config": null
1441
+ },
1442
+ {
1443
+ "pre_mixer_norm_config": {
1444
+ "epsilon": 1e-06,
1445
+ "scale_offset": 1.0,
1446
+ "upcast_mode": "only_normalization",
1447
+ "subtract_mean": false,
1448
+ "has_scale": true,
1449
+ "has_biases": false
1450
+ },
1451
+ "mixer_conv_config": null,
1452
+ "mixer_config": {
1453
+ "type": "DeltaNetConfig",
1454
+ "in_proj_config": {},
1455
+ "conv_config": {
1456
+ "has_biases": false
1457
+ },
1458
+ "out_proj_config": {},
1459
+ "norm_config": {
1460
+ "epsilon": 1e-06,
1461
+ "scale_offset": null,
1462
+ "upcast_mode": "only_normalization",
1463
+ "subtract_mean": false,
1464
+ "has_scale": true,
1465
+ "has_biases": false
1466
+ },
1467
+ "num_heads": 32,
1468
+ "num_groups": 16,
1469
+ "head_dim": 128,
1470
+ "value_head_dim": 128,
1471
+ "kernel_size": 4
1472
+ },
1473
+ "post_mixer_norm_config": null,
1474
+ "pre_mlp_norm_config": {
1475
+ "epsilon": 1e-06,
1476
+ "scale_offset": 1.0,
1477
+ "upcast_mode": "only_normalization",
1478
+ "subtract_mean": false,
1479
+ "has_scale": true,
1480
+ "has_biases": false
1481
+ },
1482
+ "mlp_conv_config": null,
1483
+ "mlp_config": {
1484
+ "type": "DenseMLPConfig",
1485
+ "linear_config": {},
1486
+ "activation": {
1487
+ "type": "SiLU",
1488
+ "alpha": 1.0
1489
+ },
1490
+ "has_up_biases": false,
1491
+ "has_down_biases": false,
1492
+ "gate_clipping": null,
1493
+ "up_clipping": null
1494
+ },
1495
+ "post_mlp_norm_config": null,
1496
+ "hidden_dim": null,
1497
+ "ple_config": null,
1498
+ "has_post_layer_scalar": false,
1499
+ "kv_source_layer_index": null,
1500
+ "rope_config": null
1501
+ },
1502
+ {
1503
+ "pre_mixer_norm_config": {
1504
+ "epsilon": 1e-06,
1505
+ "scale_offset": 1.0,
1506
+ "upcast_mode": "only_normalization",
1507
+ "subtract_mean": false,
1508
+ "has_scale": true,
1509
+ "has_biases": false
1510
+ },
1511
+ "mixer_conv_config": null,
1512
+ "mixer_config": {
1513
+ "type": "AttentionConfig",
1514
+ "qkvg_projection_config": {},
1515
+ "out_projection_config": {},
1516
+ "query_norm_config": {
1517
+ "epsilon": 1e-06,
1518
+ "scale_offset": 1.0,
1519
+ "upcast_mode": "only_normalization",
1520
+ "subtract_mean": false,
1521
+ "has_scale": true,
1522
+ "has_biases": false
1523
+ },
1524
+ "key_norm_config": {
1525
+ "epsilon": 1e-06,
1526
+ "scale_offset": 1.0,
1527
+ "upcast_mode": "only_normalization",
1528
+ "subtract_mean": false,
1529
+ "has_scale": true,
1530
+ "has_biases": false
1531
+ },
1532
+ "num_heads": 16,
1533
+ "num_groups": 4,
1534
+ "head_dim": 256,
1535
+ "is_causal": true,
1536
+ "scale": null,
1537
+ "sliding_window_size": null,
1538
+ "logit_soft_cap": null,
1539
+ "has_sinks": false,
1540
+ "has_qkvg_biases": false,
1541
+ "has_out_biases": false,
1542
+ "has_gate": true,
1543
+ "normalize_values": false,
1544
+ "is_kv_sharing": false
1545
+ },
1546
+ "post_mixer_norm_config": null,
1547
+ "pre_mlp_norm_config": {
1548
+ "epsilon": 1e-06,
1549
+ "scale_offset": 1.0,
1550
+ "upcast_mode": "only_normalization",
1551
+ "subtract_mean": false,
1552
+ "has_scale": true,
1553
+ "has_biases": false
1554
+ },
1555
+ "mlp_conv_config": null,
1556
+ "mlp_config": {
1557
+ "type": "DenseMLPConfig",
1558
+ "linear_config": {},
1559
+ "activation": {
1560
+ "type": "SiLU",
1561
+ "alpha": 1.0
1562
+ },
1563
+ "has_up_biases": false,
1564
+ "has_down_biases": false,
1565
+ "gate_clipping": null,
1566
+ "up_clipping": null
1567
+ },
1568
+ "post_mlp_norm_config": null,
1569
+ "hidden_dim": null,
1570
+ "ple_config": null,
1571
+ "has_post_layer_scalar": false,
1572
+ "kv_source_layer_index": null,
1573
+ "rope_config": {
1574
+ "type": "UnscaledRoPEConfig",
1575
+ "base": 10000000.0,
1576
+ "max_sequence_length": 262144,
1577
+ "head_dim": 64
1578
+ }
1579
+ },
1580
+ {
1581
+ "pre_mixer_norm_config": {
1582
+ "epsilon": 1e-06,
1583
+ "scale_offset": 1.0,
1584
+ "upcast_mode": "only_normalization",
1585
+ "subtract_mean": false,
1586
+ "has_scale": true,
1587
+ "has_biases": false
1588
+ },
1589
+ "mixer_conv_config": null,
1590
+ "mixer_config": {
1591
+ "type": "DeltaNetConfig",
1592
+ "in_proj_config": {},
1593
+ "conv_config": {
1594
+ "has_biases": false
1595
+ },
1596
+ "out_proj_config": {},
1597
+ "norm_config": {
1598
+ "epsilon": 1e-06,
1599
+ "scale_offset": null,
1600
+ "upcast_mode": "only_normalization",
1601
+ "subtract_mean": false,
1602
+ "has_scale": true,
1603
+ "has_biases": false
1604
+ },
1605
+ "num_heads": 32,
1606
+ "num_groups": 16,
1607
+ "head_dim": 128,
1608
+ "value_head_dim": 128,
1609
+ "kernel_size": 4
1610
+ },
1611
+ "post_mixer_norm_config": null,
1612
+ "pre_mlp_norm_config": {
1613
+ "epsilon": 1e-06,
1614
+ "scale_offset": 1.0,
1615
+ "upcast_mode": "only_normalization",
1616
+ "subtract_mean": false,
1617
+ "has_scale": true,
1618
+ "has_biases": false
1619
+ },
1620
+ "mlp_conv_config": null,
1621
+ "mlp_config": {
1622
+ "type": "DenseMLPConfig",
1623
+ "linear_config": {},
1624
+ "activation": {
1625
+ "type": "SiLU",
1626
+ "alpha": 1.0
1627
+ },
1628
+ "has_up_biases": false,
1629
+ "has_down_biases": false,
1630
+ "gate_clipping": null,
1631
+ "up_clipping": null
1632
+ },
1633
+ "post_mlp_norm_config": null,
1634
+ "hidden_dim": null,
1635
+ "ple_config": null,
1636
+ "has_post_layer_scalar": false,
1637
+ "kv_source_layer_index": null,
1638
+ "rope_config": null
1639
+ },
1640
+ {
1641
+ "pre_mixer_norm_config": {
1642
+ "epsilon": 1e-06,
1643
+ "scale_offset": 1.0,
1644
+ "upcast_mode": "only_normalization",
1645
+ "subtract_mean": false,
1646
+ "has_scale": true,
1647
+ "has_biases": false
1648
+ },
1649
+ "mixer_conv_config": null,
1650
+ "mixer_config": {
1651
+ "type": "DeltaNetConfig",
1652
+ "in_proj_config": {},
1653
+ "conv_config": {
1654
+ "has_biases": false
1655
+ },
1656
+ "out_proj_config": {},
1657
+ "norm_config": {
1658
+ "epsilon": 1e-06,
1659
+ "scale_offset": null,
1660
+ "upcast_mode": "only_normalization",
1661
+ "subtract_mean": false,
1662
+ "has_scale": true,
1663
+ "has_biases": false
1664
+ },
1665
+ "num_heads": 32,
1666
+ "num_groups": 16,
1667
+ "head_dim": 128,
1668
+ "value_head_dim": 128,
1669
+ "kernel_size": 4
1670
+ },
1671
+ "post_mixer_norm_config": null,
1672
+ "pre_mlp_norm_config": {
1673
+ "epsilon": 1e-06,
1674
+ "scale_offset": 1.0,
1675
+ "upcast_mode": "only_normalization",
1676
+ "subtract_mean": false,
1677
+ "has_scale": true,
1678
+ "has_biases": false
1679
+ },
1680
+ "mlp_conv_config": null,
1681
+ "mlp_config": {
1682
+ "type": "DenseMLPConfig",
1683
+ "linear_config": {},
1684
+ "activation": {
1685
+ "type": "SiLU",
1686
+ "alpha": 1.0
1687
+ },
1688
+ "has_up_biases": false,
1689
+ "has_down_biases": false,
1690
+ "gate_clipping": null,
1691
+ "up_clipping": null
1692
+ },
1693
+ "post_mlp_norm_config": null,
1694
+ "hidden_dim": null,
1695
+ "ple_config": null,
1696
+ "has_post_layer_scalar": false,
1697
+ "kv_source_layer_index": null,
1698
+ "rope_config": null
1699
+ },
1700
+ {
1701
+ "pre_mixer_norm_config": {
1702
+ "epsilon": 1e-06,
1703
+ "scale_offset": 1.0,
1704
+ "upcast_mode": "only_normalization",
1705
+ "subtract_mean": false,
1706
+ "has_scale": true,
1707
+ "has_biases": false
1708
+ },
1709
+ "mixer_conv_config": null,
1710
+ "mixer_config": {
1711
+ "type": "DeltaNetConfig",
1712
+ "in_proj_config": {},
1713
+ "conv_config": {
1714
+ "has_biases": false
1715
+ },
1716
+ "out_proj_config": {},
1717
+ "norm_config": {
1718
+ "epsilon": 1e-06,
1719
+ "scale_offset": null,
1720
+ "upcast_mode": "only_normalization",
1721
+ "subtract_mean": false,
1722
+ "has_scale": true,
1723
+ "has_biases": false
1724
+ },
1725
+ "num_heads": 32,
1726
+ "num_groups": 16,
1727
+ "head_dim": 128,
1728
+ "value_head_dim": 128,
1729
+ "kernel_size": 4
1730
+ },
1731
+ "post_mixer_norm_config": null,
1732
+ "pre_mlp_norm_config": {
1733
+ "epsilon": 1e-06,
1734
+ "scale_offset": 1.0,
1735
+ "upcast_mode": "only_normalization",
1736
+ "subtract_mean": false,
1737
+ "has_scale": true,
1738
+ "has_biases": false
1739
+ },
1740
+ "mlp_conv_config": null,
1741
+ "mlp_config": {
1742
+ "type": "DenseMLPConfig",
1743
+ "linear_config": {},
1744
+ "activation": {
1745
+ "type": "SiLU",
1746
+ "alpha": 1.0
1747
+ },
1748
+ "has_up_biases": false,
1749
+ "has_down_biases": false,
1750
+ "gate_clipping": null,
1751
+ "up_clipping": null
1752
+ },
1753
+ "post_mlp_norm_config": null,
1754
+ "hidden_dim": null,
1755
+ "ple_config": null,
1756
+ "has_post_layer_scalar": false,
1757
+ "kv_source_layer_index": null,
1758
+ "rope_config": null
1759
+ },
1760
+ {
1761
+ "pre_mixer_norm_config": {
1762
+ "epsilon": 1e-06,
1763
+ "scale_offset": 1.0,
1764
+ "upcast_mode": "only_normalization",
1765
+ "subtract_mean": false,
1766
+ "has_scale": true,
1767
+ "has_biases": false
1768
+ },
1769
+ "mixer_conv_config": null,
1770
+ "mixer_config": {
1771
+ "type": "AttentionConfig",
1772
+ "qkvg_projection_config": {},
1773
+ "out_projection_config": {},
1774
+ "query_norm_config": {
1775
+ "epsilon": 1e-06,
1776
+ "scale_offset": 1.0,
1777
+ "upcast_mode": "only_normalization",
1778
+ "subtract_mean": false,
1779
+ "has_scale": true,
1780
+ "has_biases": false
1781
+ },
1782
+ "key_norm_config": {
1783
+ "epsilon": 1e-06,
1784
+ "scale_offset": 1.0,
1785
+ "upcast_mode": "only_normalization",
1786
+ "subtract_mean": false,
1787
+ "has_scale": true,
1788
+ "has_biases": false
1789
+ },
1790
+ "num_heads": 16,
1791
+ "num_groups": 4,
1792
+ "head_dim": 256,
1793
+ "is_causal": true,
1794
+ "scale": null,
1795
+ "sliding_window_size": null,
1796
+ "logit_soft_cap": null,
1797
+ "has_sinks": false,
1798
+ "has_qkvg_biases": false,
1799
+ "has_out_biases": false,
1800
+ "has_gate": true,
1801
+ "normalize_values": false,
1802
+ "is_kv_sharing": false
1803
+ },
1804
+ "post_mixer_norm_config": null,
1805
+ "pre_mlp_norm_config": {
1806
+ "epsilon": 1e-06,
1807
+ "scale_offset": 1.0,
1808
+ "upcast_mode": "only_normalization",
1809
+ "subtract_mean": false,
1810
+ "has_scale": true,
1811
+ "has_biases": false
1812
+ },
1813
+ "mlp_conv_config": null,
1814
+ "mlp_config": {
1815
+ "type": "DenseMLPConfig",
1816
+ "linear_config": {},
1817
+ "activation": {
1818
+ "type": "SiLU",
1819
+ "alpha": 1.0
1820
+ },
1821
+ "has_up_biases": false,
1822
+ "has_down_biases": false,
1823
+ "gate_clipping": null,
1824
+ "up_clipping": null
1825
+ },
1826
+ "post_mlp_norm_config": null,
1827
+ "hidden_dim": null,
1828
+ "ple_config": null,
1829
+ "has_post_layer_scalar": false,
1830
+ "kv_source_layer_index": null,
1831
+ "rope_config": {
1832
+ "type": "UnscaledRoPEConfig",
1833
+ "base": 10000000.0,
1834
+ "max_sequence_length": 262144,
1835
+ "head_dim": 64
1836
+ }
1837
+ },
1838
+ {
1839
+ "pre_mixer_norm_config": {
1840
+ "epsilon": 1e-06,
1841
+ "scale_offset": 1.0,
1842
+ "upcast_mode": "only_normalization",
1843
+ "subtract_mean": false,
1844
+ "has_scale": true,
1845
+ "has_biases": false
1846
+ },
1847
+ "mixer_conv_config": null,
1848
+ "mixer_config": {
1849
+ "type": "DeltaNetConfig",
1850
+ "in_proj_config": {},
1851
+ "conv_config": {
1852
+ "has_biases": false
1853
+ },
1854
+ "out_proj_config": {},
1855
+ "norm_config": {
1856
+ "epsilon": 1e-06,
1857
+ "scale_offset": null,
1858
+ "upcast_mode": "only_normalization",
1859
+ "subtract_mean": false,
1860
+ "has_scale": true,
1861
+ "has_biases": false
1862
+ },
1863
+ "num_heads": 32,
1864
+ "num_groups": 16,
1865
+ "head_dim": 128,
1866
+ "value_head_dim": 128,
1867
+ "kernel_size": 4
1868
+ },
1869
+ "post_mixer_norm_config": null,
1870
+ "pre_mlp_norm_config": {
1871
+ "epsilon": 1e-06,
1872
+ "scale_offset": 1.0,
1873
+ "upcast_mode": "only_normalization",
1874
+ "subtract_mean": false,
1875
+ "has_scale": true,
1876
+ "has_biases": false
1877
+ },
1878
+ "mlp_conv_config": null,
1879
+ "mlp_config": {
1880
+ "type": "DenseMLPConfig",
1881
+ "linear_config": {},
1882
+ "activation": {
1883
+ "type": "SiLU",
1884
+ "alpha": 1.0
1885
+ },
1886
+ "has_up_biases": false,
1887
+ "has_down_biases": false,
1888
+ "gate_clipping": null,
1889
+ "up_clipping": null
1890
+ },
1891
+ "post_mlp_norm_config": null,
1892
+ "hidden_dim": null,
1893
+ "ple_config": null,
1894
+ "has_post_layer_scalar": false,
1895
+ "kv_source_layer_index": null,
1896
+ "rope_config": null
1897
+ },
1898
+ {
1899
+ "pre_mixer_norm_config": {
1900
+ "epsilon": 1e-06,
1901
+ "scale_offset": 1.0,
1902
+ "upcast_mode": "only_normalization",
1903
+ "subtract_mean": false,
1904
+ "has_scale": true,
1905
+ "has_biases": false
1906
+ },
1907
+ "mixer_conv_config": null,
1908
+ "mixer_config": {
1909
+ "type": "DeltaNetConfig",
1910
+ "in_proj_config": {},
1911
+ "conv_config": {
1912
+ "has_biases": false
1913
+ },
1914
+ "out_proj_config": {},
1915
+ "norm_config": {
1916
+ "epsilon": 1e-06,
1917
+ "scale_offset": null,
1918
+ "upcast_mode": "only_normalization",
1919
+ "subtract_mean": false,
1920
+ "has_scale": true,
1921
+ "has_biases": false
1922
+ },
1923
+ "num_heads": 32,
1924
+ "num_groups": 16,
1925
+ "head_dim": 128,
1926
+ "value_head_dim": 128,
1927
+ "kernel_size": 4
1928
+ },
1929
+ "post_mixer_norm_config": null,
1930
+ "pre_mlp_norm_config": {
1931
+ "epsilon": 1e-06,
1932
+ "scale_offset": 1.0,
1933
+ "upcast_mode": "only_normalization",
1934
+ "subtract_mean": false,
1935
+ "has_scale": true,
1936
+ "has_biases": false
1937
+ },
1938
+ "mlp_conv_config": null,
1939
+ "mlp_config": {
1940
+ "type": "DenseMLPConfig",
1941
+ "linear_config": {},
1942
+ "activation": {
1943
+ "type": "SiLU",
1944
+ "alpha": 1.0
1945
+ },
1946
+ "has_up_biases": false,
1947
+ "has_down_biases": false,
1948
+ "gate_clipping": null,
1949
+ "up_clipping": null
1950
+ },
1951
+ "post_mlp_norm_config": null,
1952
+ "hidden_dim": null,
1953
+ "ple_config": null,
1954
+ "has_post_layer_scalar": false,
1955
+ "kv_source_layer_index": null,
1956
+ "rope_config": null
1957
+ },
1958
+ {
1959
+ "pre_mixer_norm_config": {
1960
+ "epsilon": 1e-06,
1961
+ "scale_offset": 1.0,
1962
+ "upcast_mode": "only_normalization",
1963
+ "subtract_mean": false,
1964
+ "has_scale": true,
1965
+ "has_biases": false
1966
+ },
1967
+ "mixer_conv_config": null,
1968
+ "mixer_config": {
1969
+ "type": "DeltaNetConfig",
1970
+ "in_proj_config": {},
1971
+ "conv_config": {
1972
+ "has_biases": false
1973
+ },
1974
+ "out_proj_config": {},
1975
+ "norm_config": {
1976
+ "epsilon": 1e-06,
1977
+ "scale_offset": null,
1978
+ "upcast_mode": "only_normalization",
1979
+ "subtract_mean": false,
1980
+ "has_scale": true,
1981
+ "has_biases": false
1982
+ },
1983
+ "num_heads": 32,
1984
+ "num_groups": 16,
1985
+ "head_dim": 128,
1986
+ "value_head_dim": 128,
1987
+ "kernel_size": 4
1988
+ },
1989
+ "post_mixer_norm_config": null,
1990
+ "pre_mlp_norm_config": {
1991
+ "epsilon": 1e-06,
1992
+ "scale_offset": 1.0,
1993
+ "upcast_mode": "only_normalization",
1994
+ "subtract_mean": false,
1995
+ "has_scale": true,
1996
+ "has_biases": false
1997
+ },
1998
+ "mlp_conv_config": null,
1999
+ "mlp_config": {
2000
+ "type": "DenseMLPConfig",
2001
+ "linear_config": {},
2002
+ "activation": {
2003
+ "type": "SiLU",
2004
+ "alpha": 1.0
2005
+ },
2006
+ "has_up_biases": false,
2007
+ "has_down_biases": false,
2008
+ "gate_clipping": null,
2009
+ "up_clipping": null
2010
+ },
2011
+ "post_mlp_norm_config": null,
2012
+ "hidden_dim": null,
2013
+ "ple_config": null,
2014
+ "has_post_layer_scalar": false,
2015
+ "kv_source_layer_index": null,
2016
+ "rope_config": null
2017
+ },
2018
+ {
2019
+ "pre_mixer_norm_config": {
2020
+ "epsilon": 1e-06,
2021
+ "scale_offset": 1.0,
2022
+ "upcast_mode": "only_normalization",
2023
+ "subtract_mean": false,
2024
+ "has_scale": true,
2025
+ "has_biases": false
2026
+ },
2027
+ "mixer_conv_config": null,
2028
+ "mixer_config": {
2029
+ "type": "AttentionConfig",
2030
+ "qkvg_projection_config": {},
2031
+ "out_projection_config": {},
2032
+ "query_norm_config": {
2033
+ "epsilon": 1e-06,
2034
+ "scale_offset": 1.0,
2035
+ "upcast_mode": "only_normalization",
2036
+ "subtract_mean": false,
2037
+ "has_scale": true,
2038
+ "has_biases": false
2039
+ },
2040
+ "key_norm_config": {
2041
+ "epsilon": 1e-06,
2042
+ "scale_offset": 1.0,
2043
+ "upcast_mode": "only_normalization",
2044
+ "subtract_mean": false,
2045
+ "has_scale": true,
2046
+ "has_biases": false
2047
+ },
2048
+ "num_heads": 16,
2049
+ "num_groups": 4,
2050
+ "head_dim": 256,
2051
+ "is_causal": true,
2052
+ "scale": null,
2053
+ "sliding_window_size": null,
2054
+ "logit_soft_cap": null,
2055
+ "has_sinks": false,
2056
+ "has_qkvg_biases": false,
2057
+ "has_out_biases": false,
2058
+ "has_gate": true,
2059
+ "normalize_values": false,
2060
+ "is_kv_sharing": false
2061
+ },
2062
+ "post_mixer_norm_config": null,
2063
+ "pre_mlp_norm_config": {
2064
+ "epsilon": 1e-06,
2065
+ "scale_offset": 1.0,
2066
+ "upcast_mode": "only_normalization",
2067
+ "subtract_mean": false,
2068
+ "has_scale": true,
2069
+ "has_biases": false
2070
+ },
2071
+ "mlp_conv_config": null,
2072
+ "mlp_config": {
2073
+ "type": "DenseMLPConfig",
2074
+ "linear_config": {},
2075
+ "activation": {
2076
+ "type": "SiLU",
2077
+ "alpha": 1.0
2078
+ },
2079
+ "has_up_biases": false,
2080
+ "has_down_biases": false,
2081
+ "gate_clipping": null,
2082
+ "up_clipping": null
2083
+ },
2084
+ "post_mlp_norm_config": null,
2085
+ "hidden_dim": null,
2086
+ "ple_config": null,
2087
+ "has_post_layer_scalar": false,
2088
+ "kv_source_layer_index": null,
2089
+ "rope_config": {
2090
+ "type": "UnscaledRoPEConfig",
2091
+ "base": 10000000.0,
2092
+ "max_sequence_length": 262144,
2093
+ "head_dim": 64
2094
+ }
2095
+ }
2096
+ ],
2097
+ "output_norm_config": {
2098
+ "epsilon": 1e-06,
2099
+ "scale_offset": 1.0,
2100
+ "upcast_mode": "only_normalization",
2101
+ "subtract_mean": false,
2102
+ "has_scale": true,
2103
+ "has_biases": false
2104
+ },
2105
+ "model_dim": 4096,
2106
+ "hidden_dim": 12288
2107
+ },
2108
+ "vocab_size": 248320,
2109
+ "ple_model_config": null,
2110
+ "embedding_norm_config": null
2111
+ },
2112
+ "generation_config": {
2113
+ "stop_token_ids": [
2114
+ 248044,
2115
+ 248046
2116
+ ],
2117
+ "temperature": 0.6,
2118
+ "top_k": 20,
2119
+ "top_p": 0.95,
2120
+ "min_p": null,
2121
+ "banned_tokens": null,
2122
+ "repetition_penalty": null,
2123
+ "presence_penalty": null,
2124
+ "frequency_penalty": null,
2125
+ "suffix_repetition_length": null
2126
+ }
2127
+ }
encoding.json ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ {
2
+ "type": "hanashi",
3
+ "name": "qwen3.5"
4
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:16535f7ef92f1828a4270eb0f0ffe30fb33359521fa672439786d2c59c3d6517
3
+ size 5189118800
provenance.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "kind": "rtn",
3
+ "source": "huihui-ai/Huihui-Qwen3.5-9B-Claude-4.6-Opus-abliterated",
4
+ "preconditioners": null,
5
+ "calibration_json": null,
6
+ "calibration": null,
7
+ "overlap_checked": true,
8
+ "lalamo_commit": "e401e17761f2fcd337b0cf20218d67d656e6ecba",
9
+ "uzu_commit": "fa0aa23d",
10
+ "created": "2026-11-07"
11
+ }
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:87a7830d63fcf43bf241c3c5242e96e62dd3fdc29224ca26fed8ea333db72de4
3
+ size 19989343