Hugging Face's logo

amd
/
DeepSeek-R1-0528-MXFP4

Safetensors
deepseek_v3
custom_code
8-bit precision
quark
Model card Files
xet
Community
3
linzhao-amd commited on
Commit
8726bd7
·
verified ·
1 Parent(s): d53b0ef

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +1 -1
README.md CHANGED
@@ -100,7 +100,7 @@ The model was evaluated on AIME24, GPQA Diamond, and MATH-500 benchmarks using t
100
 
101
  ### Reproduction
102
 
103
- The results can be reproduced using vLLM emulation docker `rocmshared/pytorch:vllm-gfx950-mxfp4-mxfp6-v3` and [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) evaluation framwork. We took 10 rounds with different random seeds for reliable performance.
104
 
105
  ```
106
  MODEL_ARGS="model_name=amd/DeepSeek-R1-0528-MXFP4-Preview,dtype=bfloat16,tensor_parallel_size=8,max_model_length=71536,max_num_batched_tokens=32768,gpu_memory_utilization=0.85,generation_parameters={max_new_tokens:65536,temperature:0.6,top_p:0.95,seed:$SEED}"
 
100
 
101
  ### Reproduction
102
 
103
+ The results of AIME24, MATH-500, and GPQA Diamond, were obtained using [lighteval](https://github.com/huggingface/lighteval/tree/v0.10.0) and [vLLM](https://docs.vllm.ai/en/latest/), running with docker `rocm/vllm-private:pytorch-vllm-gfx950-mxfp4-mxfp6-v3`.
104
 
105
  ```
106
  MODEL_ARGS="model_name=amd/DeepSeek-R1-0528-MXFP4-Preview,dtype=bfloat16,tensor_parallel_size=8,max_model_length=71536,max_num_batched_tokens=32768,gpu_memory_utilization=0.85,generation_parameters={max_new_tokens:65536,temperature:0.6,top_p:0.95,seed:$SEED}"