3.06 GB

Ctrl+K

1 contributor

History: 15 commits

Koopah

Document eval_throughput.py usage

815f706 verified 3 months ago

docker
DSPARK draft for unsloth/Qwen3.6-35B-A3B-NVFP4: weights, model card, benchmarks, Dockerfile (training in progress) 4 months ago
.gitattributes

1.74 kB
Rebench under sglang-blog protocol (one-batch, fixed concat-16q prompt, OSL1024, 3 rounds): new charts + tables both engines, k-ladder as confidence-scheduler simulation, -Fast variant section 4 months ago
README.md

17.6 kB
Document eval_throughput.py usage 3 months ago
config.json

1.36 kB
config: add dflash_config sub-dict for vLLM compatibility (reads target_layer_ids/mask_token_id from there) 4 months ago
eval_throughput.py

8.75 kB
Add portable throughput/chart eval script 3 months ago
frontier_fast_vllm.png

144 kB
xet

Rebench under sglang-blog protocol (one-batch, fixed concat-16q prompt, OSL1024, 3 rounds): new charts + tables both engines, k-ladder as confidence-scheduler simulation, -Fast variant section 4 months ago
frontier_gsm8k.png

149 kB
xet

sglang chart: add MTP k=4 arm; restore continuous-serving (closed-loop) section with its chart 4 months ago
frontier_sglang.png

206 kB
xet

sglang chart: add MTP k=4 arm; restore continuous-serving (closed-loop) section with its chart 4 months ago
frontier_vllm.png

186 kB
xet

Rebench under sglang-blog protocol (one-batch, fixed concat-16q prompt, OSL1024, 3 rounds): new charts + tables both engines, k-ladder as confidence-scheduler simulation, -Fast variant section 4 months ago
model.safetensors

3.06 GB
xet

DSPARK draft for unsloth/Qwen3.6-35B-A3B-NVFP4: weights, model card, benchmarks, Dockerfile (training in progress) 4 months ago