Add GPQA Diamond and fix evaluation deps (huggingface#196)

* Add GPQA Diamond * Add table * Fix README * Up * Fixes * Ignore logs * Fix * Pin deps * Fix GRPO * Add Llama 70B tabels * Restore dp * Pin lighteval * Use bfloat16 * Tune table * Add note
GitMonkey0 · Feb 6, 2025 · 5c5505d · 5c5505d
1 parent 5f8d947
commit 5c5505d
Show file tree

Hide file tree

Showing 9 changed files with 202 additions and 148 deletions.
diff --git a/Makefile b/Makefile
@@ -26,7 +26,7 @@ evaluate:
 		fi \
 	),))
 	$(if $(filter tensor,$(PARALLEL)),export VLLM_WORKER_MULTIPROC_METHOD=spawn &&,) \
-	MODEL_ARGS="pretrained=$(MODEL),dtype=float16,$(PARALLEL_ARGS),max_model_length=32768,gpu_memory_utilisation=0.8" && \
+	MODEL_ARGS="pretrained=$(MODEL),dtype=bfloat16,$(PARALLEL_ARGS),max_model_length=32768,gpu_memory_utilisation=0.8" && \
 	lighteval vllm $$MODEL_ARGS "custom|$(TASK)|0|0" \
 		--custom-tasks src/open_r1/evaluate.py \
 		--use-chat-template \

diff --git a/README.md b/README.md
@@ -50,23 +50,23 @@ To install `uv`, follow the [UV Installation Guide](https://docs.astral.sh/uv/ge
 
 
 ```shell
-uv venv openr1 --python 3.11 && source openr1/bin/activate && uv pip install --upgrade pip
+uv venv openr1 --python 3.11 && source openr1/bin/activate && uv pip install --upgrade pip --link-mode=copy
 ```
 
 Next, install vLLM:
 
 ```shell
-uv pip install vllm>=0.7.0
+uv pip install vllm==0.7.1
 
 # For CUDA 12.1
-pip install vllm>=0.7.0 --extra-index-url https://download.pytorch.org/whl/cu121
+uv pip install vllm==0.7.1 --extra-index-url https://download.pytorch.org/whl/cu121 --index-strategy unsafe-best-match --link-mode=copy
 export LD_LIBRARY_PATH=$(python -c "import site; print(site.getsitepackages()[0] + '/nvidia/nvjitlink/lib')"):$LD_LIBRARY_PATH
 ```
 
 This will also install PyTorch `v2.5.1` and it is **very important** to use this version since the vLLM binaries are compiled for it. You can then install the remaining dependencies for your specific use case via `pip install -e .[LIST OF MODES]`. For most contributors, we recommend:
 
 ```shell
-pip install -e ".[dev]"
+GIT_LFS_SKIP_SMUDGE=1 uv pip install -e ".[dev]" --link-mode=copy
 ```
 
 Next, log into your Hugging Face and Weights and Biases accounts as follows:
@@ -141,30 +141,46 @@ We use `lighteval` to evaluate models, with custom tasks defined in `src/open_r1
 
 ```shell
 MODEL=deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
-MODEL_ARGS="pretrained=$MODEL,dtype=float16,max_model_length=32768,gpu_memory_utilisation=0.8"
-TASK=aime24
+MODEL_ARGS="pretrained=$MODEL,dtype=bfloat16,max_model_length=32768,gpu_memory_utilisation=0.8"
 OUTPUT_DIR=data/evals/$MODEL
 
+# AIME 2024
+TASK=aime24
+lighteval vllm $MODEL_ARGS "custom|$TASK|0|0" \
+    --custom-tasks src/open_r1/evaluate.py \
+    --use-chat-template \
+    --output-dir $OUTPUT_DIR
+
+# MATH-500
+TASK=math_500
+lighteval vllm $MODEL_ARGS "custom|$TASK|0|0" \
+    --custom-tasks src/open_r1/evaluate.py \
+    --use-chat-template \
+    --output-dir $OUTPUT_DIR
+
+# GPQA Diamond
+TASK=gpqa:diamond
 lighteval vllm $MODEL_ARGS "custom|$TASK|0|0" \
     --custom-tasks src/open_r1/evaluate.py \
     --use-chat-template \
-    --system-prompt="Please reason step by step, and put your final answer within \boxed{}." \
     --output-dir $OUTPUT_DIR 
 ```
 
+> [!IMPORTANT]
+> You must set `max_model_length=32768` in the `vllm` command to align with the `generation_size` we define per eval. Without this, `lighteval` will throw an error.
+
 To increase throughput across multiple GPUs, use _data parallel_ as follows:
 
 ```shell
 NUM_GPUS=8
 MODEL=deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
-MODEL_ARGS="pretrained=$MODEL,dtype=float16,data_parallel_size=$NUM_GPUS,max_model_length=32768,gpu_memory_utilisation=0.8"
+MODEL_ARGS="pretrained=$MODEL,dtype=bfloat16,data_parallel_size=$NUM_GPUS,max_model_length=32768,gpu_memory_utilisation=0.8"
 TASK=aime24
 OUTPUT_DIR=data/evals/$MODEL
 
 lighteval vllm $MODEL_ARGS "custom|$TASK|0|0" \
     --custom-tasks src/open_r1/evaluate.py \
     --use-chat-template \
-    --system-prompt="Please reason step by step, and put your final answer within \boxed{}." \
     --output-dir $OUTPUT_DIR 
 ```
 
@@ -173,58 +189,105 @@ For large models which require sharding across GPUs, use _tensor parallel_ and r
 ```shell
 NUM_GPUS=8
 MODEL=deepseek-ai/DeepSeek-R1-Distill-Qwen-32B
-MODEL_ARGS="pretrained=$MODEL,dtype=float16,tensor_parallel_size=$NUM_GPUS,max_model_length=32768,gpu_memory_utilisation=0.8"
+MODEL_ARGS="pretrained=$MODEL,dtype=bfloat16,tensor_parallel_size=$NUM_GPUS,max_model_length=32768,gpu_memory_utilisation=0.8"
 TASK=aime24
 OUTPUT_DIR=data/evals/$MODEL
 
 export VLLM_WORKER_MULTIPROC_METHOD=spawn
 lighteval vllm $MODEL_ARGS "custom|$TASK|0|0" \
     --custom-tasks src/open_r1/evaluate.py \
     --use-chat-template \
-    --system-prompt="Please reason step by step, and put your final answer within \boxed{}." \
     --output-dir $OUTPUT_DIR 
 ```
 
 You can also launch an evaluation with `make evaluate`, specifying the model, task, and optionally the parallelism technique and number of GPUs.
 
 To evaluate on a single GPU:
+
 ```shell
 make evaluate MODEL=deepseek-ai/DeepSeek-R1-Distill-Qwen-32B TASK=aime24
 ```
 
 To use Data Parallelism:
+
 ```shell
 make evaluate MODEL=deepseek-ai/DeepSeek-R1-Distill-Qwen-32B TASK=aime24 PARALLEL=data NUM_GPUS=8
 ```
 
 To use Tensor Parallelism:
+
 ```shell
 make evaluate MODEL=deepseek-ai/DeepSeek-R1-Distill-Qwen-32B TASK=aime24 PARALLEL=tensor NUM_GPUS=8
 ```
-## Reproducing Deepseek's evaluation results on MATH-500
-We are able to reproduce Deepseek's reported results on the MATH-500 Benchmark:
-| Model                      | MATH-500 (HF lighteval) | MATH-500 (DeepSeek Reported) |
-| :-------------------------- | :-------: | :----------------------------: |
-| DeepSeek-R1-Distill-Qwen-1.5B  |  81.6   |              83.9              |
-| DeepSeek-R1-Distill-Qwen-7B    |  91.8   |              92.8              |
-| DeepSeek-R1-Distill-Qwen-14B   |  94.2   |              93.9              |
-| DeepSeek-R1-Distill-Qwen-32B   |  95.0   |              94.3              |
-| DeepSeek-R1-Distill-Llama-8B   |  85.8   |              89.1              |
-| DeepSeek-R1-Distill-Llama-70B  |  93.4   |              94.5              |
 
+## Reproducing Deepseek's evaluation results
+
+> [!NOTE]
+> The DeepSeek-R1 paper uses sampling with a temperature of 0.6, a top-p value of 0.95, and 64 responses per query to estimate `pass@1`. Below, we report the results from greedy decoding, which likely explains the small 1-3σ discrepancies between our results and theirs.
+
+### MATH-500
+
+We are able to reproduce Deepseek's reported results on the MATH-500 benchmark within ~1-3 standard deviations:
 
+| Model                         | MATH-500 (🤗 LightEval) | MATH-500 (DeepSeek Reported) |
+|:------------------------------|:-----------------------:|:----------------------------:|
+| DeepSeek-R1-Distill-Qwen-1.5B |          81.2           |             83.9             |
+| DeepSeek-R1-Distill-Qwen-7B   |          91.8           |             92.8             |
+| DeepSeek-R1-Distill-Qwen-14B  |          94.2           |             93.9             |
+| DeepSeek-R1-Distill-Qwen-32B  |          95.0           |             94.3             |
+| DeepSeek-R1-Distill-Llama-8B  |          85.4           |             89.1             |
+| DeepSeek-R1-Distill-Llama-70B |          93.4           |             94.5             |
 
 To reproduce these results use the following command:
+
+```shell
+NUM_GPUS=1 # Set to 8 for 32B and 70B models
+MODEL=deepseek-ai/{model_name}
+MODEL_ARGS="pretrained=$MODEL,dtype=bfloat16,max_model_length=32768,gpu_memory_utilisation=0.8,tensor_parallel_size=$NUM_GPUS"
+OUTPUT_DIR=data/evals/$MODEL
+
+lighteval vllm $MODEL_ARGS "custom|math_500|0|0" \
+    --custom-tasks src/open_r1/evaluate.py \
+    --use-chat-template \
+    --output-dir $OUTPUT_DIR
+```
+
+Alternatively, you can launch Slurm jobs as follows:
+
 ```shell
-sbatch slurm/evaluate.slurm deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B math_500
-sbatch slurm/evaluate.slurm deepseek-ai/DeepSeek-R1-Distill-Qwen-7B math_500
-sbatch slurm/evaluate.slurm deepseek-ai/DeepSeek-R1-Distill-Qwen-14B math_500
-sbatch slurm/evaluate.slurm deepseek-ai/DeepSeek-R1-Distill-Qwen-32B math_500 tp
-sbatch slurm/evaluate.slurm deepseek-ai/DeepSeek-R1-Distill-Llama-8B math_500
-sbatch slurm/evaluate.slurm deepseek-ai/DeepSeek-R1-Distill-Llama-70B math_500 tp
+python scripts/run_benchmarks.py --model-id={model_id}  --benchmarks math_500
 ```
 
+### GPQA Diamond
+
+We are able to reproduce Deepseek's reported results on the GPQA Diamond benchmark within ~1-3 standard deviations:
+
+| Model                         | GPQA Diamond (🤗 LightEval) | GPQA Diamond (DeepSeek Reported) |
+|:------------------------------|:---------------------------:|:--------------------------------:|
+| DeepSeek-R1-Distill-Qwen-1.5B |            33.3             |               33.8               |
+| DeepSeek-R1-Distill-Qwen-7B   |            48.4             |               49.1               |
+| DeepSeek-R1-Distill-Qwen-14B  |            55.6             |               59.1               |
+| DeepSeek-R1-Distill-Qwen-32B  |            58.6             |               62.1               |
+| DeepSeek-R1-Distill-Llama-8B  |            51.0             |               49.0               |
+| DeepSeek-R1-Distill-Llama-70B |            65.2             |               65.2               |
+
+To reproduce these results use the following command:
+
+```shell
+NUM_GPUS=1 # Set to 8 for 32B and 70B models
+MODEL=deepseek-ai/{model_name}
+MODEL_ARGS="pretrained=$MODEL,dtype=bfloat16,max_model_length=32768,gpu_memory_utilisation=0.8,tensor_parallel_size=$NUM_GPUS"
+OUTPUT_DIR=data/evals/$MODEL
+
+lighteval vllm $MODEL_ARGS "custom|gpqa:diamond|0|0" \
+    --custom-tasks src/open_r1/evaluate.py \
+    --use-chat-template \
+    --output-dir $OUTPUT_DIR
+```
 
+```shell
+python scripts/run_benchmarks.py --model-id={model_id}  --benchmarks gpqa
+```
 
 ## Data generation
 

diff --git a/logs/.gitkeep b/logs/.gitkeep
diff --git a/setup.py b/setup.py
@@ -53,17 +53,17 @@
     "huggingface-hub[cli]>=0.19.2,<1.0",
     "isort>=5.12.0",
     "liger_kernel==0.5.2",
-    "lighteval @ git+https://github.com/huggingface/lighteval.git@0e462692436e1f0575bdb4c6ef63453ad9bde7d4#egg=lighteval[math]",
-    "math-verify>=0.3.3",  # Used for math verification in grpo
+    "lighteval @ git+https://github.com/huggingface/lighteval.git@86f62259f105ae164f655e0b91c92a823a742724#egg=lighteval[math]",
+    "math-verify==0.5.2",  # Used for math verification in grpo
     "packaging>=23.0",
     "parameterized>=0.9.0",
     "pytest",
     "safetensors>=0.3.3",
     "sentencepiece>=0.1.99",
-    "torch>=2.5.1",
+    "torch==2.5.1",
     "transformers @ git+https://github.com/huggingface/transformers.git@main",
     "trl @ git+https://github.com/huggingface/trl.git@main",
-    "vllm>=0.7.1",
+    "vllm==0.7.1",
     "wandb>=0.19.1",
 ]
 

diff --git a/slurm/eval_callback.slurm b/slurm/eval_callback.slurm