Upload folder using huggingface_hub

README.md

@@ -0,0 +1,48 @@
+---
+license: apache-2.0
+language:
+- pl
+base_model:
+- speakleash/Bielik-4.5B-v3.0-Instruct
+pipeline_tag: text-generation
+library_name: transformers
+tags:
+- bielik
+- polish
+- nvfp4
+- fp4
+- compressed-tensors
+- vllm
+- quantized
+---
+
+# Bielik-4.5B-v3.0-Instruct — NVFP4 (W4A4)
+
+Kwantyzacja **NVFP4 (W4A4)** modelu [speakleash/Bielik-4.5B-v3.0-Instruct](https://huggingface.co/speakleash/Bielik-4.5B-v3.0-Instruct),
+w formacie `compressed-tensors` (`nvfp4-pack-quantized`) gotowym do **vLLM**.
+
+- **Wagi:** 4-bit (FP4 E2M1), grupy po 16, blokowe skale FP8 (E4M3) + globalna skala FP32
+- **Aktywacje:** 4-bit FP4 (dynamiczne, per-grupa)
+- **Pominięte:** `lm_head` (pełna precyzja)
+- **Rozmiar:** ~8,9 GB (bf16) → **~2,7 GB**
+
+## Kalibracja
+
+PTQ przez [`llm-compressor`](https://github.com/vllm-project/llm-compressor), 512 próbek polskiej Wikipedii,
+długość sekwencji 2048. Kalibracja na polskim tekście dla zachowania jakości w języku docelowym.
+
+## Sprzęt
+
+NVFP4 jest akcelerowany sprzętowo na **Blackwellu (sm_100+)**. Na starszych GPU (np. Ada/RTX 4090) vLLM
+ładuje model przez kernel **Marlin (weight-only FP4)** — działa wszędzie, zysk to mniejszy VRAM.
+
+## Użycie (vLLM)
+
+```bash
+vllm serve <repo>/Bielik-4.5B-v3.0-Instruct-NVFP4 --max-model-len 8192
+```
+
+## Licencja i atrybucja
+
+Apache-2.0, dziedziczona z modelu bazowego (SpeakLeash & ACK Cyfronet AGH).
+Ten checkpoint to wyłącznie kwantyzacja oryginalnych wag.

chat_template.jinja

@@ -0,0 +1,4 @@
+{{bos_token}}{% for message in messages %}{{'<|im_start|>' + message['role'] + '
+' + message['content'] + '<|im_end|>' + '
+'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant
+' }}{% endif %}

config.json

@@ -0,0 +1,83 @@
+{
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": true,
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "dtype": "bfloat16",
+  "eos_token_id": [
+    4,
+    2
+  ],
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_position_embeddings": 32768,
+  "mlp_bias": true,
+  "model_type": "llama",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 60,
+  "num_key_value_heads": 2,
+  "pad_token_id": 2,
+  "pretraining_tp": 1,
+  "quantization_config": {
+    "config_groups": {
+      "group_0": {
+        "format": "nvfp4-pack-quantized",
+        "input_activations": {
+          "actorder": null,
+          "block_structure": null,
+          "dynamic": "local",
+          "group_size": 16,
+          "num_bits": 4,
+          "observer": "static_minmax",
+          "observer_kwargs": {},
+          "scale_dtype": "torch.float8_e4m3fn",
+          "strategy": "tensor_group",
+          "symmetric": true,
+          "type": "float",
+          "zp_dtype": null
+        },
+        "output_activations": null,
+        "targets": [
+          "Linear"
+        ],
+        "weights": {
+          "actorder": null,
+          "block_structure": null,
+          "dynamic": false,
+          "group_size": 16,
+          "num_bits": 4,
+          "observer": "memoryless_minmax",
+          "observer_kwargs": {},
+          "scale_dtype": "torch.float8_e4m3fn",
+          "strategy": "tensor_group",
+          "symmetric": true,
+          "type": "float",
+          "zp_dtype": null
+        }
+      }
+    },
+    "format": "nvfp4-pack-quantized",
+    "global_compression_ratio": null,
+    "ignore": [
+      "lm_head"
+    ],
+    "kv_cache_scheme": null,
+    "quant_method": "compressed-tensors",
+    "quantization_status": "compressed",
+    "sparsity_config": {},
+    "transform_config": {},
+    "version": "0.14.0.1"
+  },
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000,
+  "tie_word_embeddings": false,
+  "transformers_version": "4.57.6",
+  "use_cache": true,
+  "vocab_size": 32000
+}

generation_config.json

@@ -0,0 +1,10 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": [
+    4,
+    2
+  ],
+  "pad_token_id": 2,
+  "transformers_version": "4.57.6"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ade2e3c74ba846f4f4a5f7aa9fa1b31874bd2b9dc1cc34fec82f8c63039e7966
+size 2867458640

recipe.yaml

@@ -0,0 +1,7 @@
+default_stage:
+  default_modifiers:
+    QuantizationModifier:
+      targets: [Linear]
+      ignore: [lm_head]
+      scheme: NVFP4
+      bypass_divisibility_checks: false

special_tokens_map.json

@@ -0,0 +1,36 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<tool_call>",
+    "</tool_call>"
+  ],
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer_config.json

@@ -0,0 +1,81 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": null,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "3": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "4": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "5": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "6": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<tool_call>",
+    "</tool_call>"
+  ],
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "extra_special_tokens": {},
+  "legacy": true,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "</s>",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizerFast",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}