Instructions to use RedHatAI/Meta-Llama-3.1-70B-Instruct-FP8-dynamic with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use RedHatAI/Meta-Llama-3.1-70B-Instruct-FP8-dynamic with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="RedHatAI/Meta-Llama-3.1-70B-Instruct-FP8-dynamic")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("RedHatAI/Meta-Llama-3.1-70B-Instruct-FP8-dynamic")
model = AutoModelForCausalLM.from_pretrained("RedHatAI/Meta-Llama-3.1-70B-Instruct-FP8-dynamic")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use RedHatAI/Meta-Llama-3.1-70B-Instruct-FP8-dynamic with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "RedHatAI/Meta-Llama-3.1-70B-Instruct-FP8-dynamic"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "RedHatAI/Meta-Llama-3.1-70B-Instruct-FP8-dynamic",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/RedHatAI/Meta-Llama-3.1-70B-Instruct-FP8-dynamic

SGLang

How to use RedHatAI/Meta-Llama-3.1-70B-Instruct-FP8-dynamic with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "RedHatAI/Meta-Llama-3.1-70B-Instruct-FP8-dynamic" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "RedHatAI/Meta-Llama-3.1-70B-Instruct-FP8-dynamic",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "RedHatAI/Meta-Llama-3.1-70B-Instruct-FP8-dynamic" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "RedHatAI/Meta-Llama-3.1-70B-Instruct-FP8-dynamic",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use RedHatAI/Meta-Llama-3.1-70B-Instruct-FP8-dynamic with Docker Model Runner:
```
docker model run hf.co/RedHatAI/Meta-Llama-3.1-70B-Instruct-FP8-dynamic
```

Lin-K76 commited on Aug 23, 2024

Commit

019d944

verified ·

1 Parent(s): 17e44d2

Update README.md

Browse files

Files changed (1) hide show

README.md +47 -26

README.md CHANGED Viewed

@@ -33,7 +33,7 @@ base_model: meta-llama/Meta-Llama-3.1-70B-Instruct
 - **Model Developers:** Neural Magic
 Quantized version of [Meta-Llama-3.1-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-70B-Instruct).
-It achieves an average score of 83.41 on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) benchmark (version 1), whereas the unquantized model achieves 83.61.
 ### Model Optimizations
@@ -118,11 +118,11 @@ model_stub = "meta-llama/Meta-Llama-3.1-70B-Instruct"
 model_name = model_stub.split("/")[-1]
 device_map = calculate_offload_device_map(
-    model_stub, reserve_for_hessians=False, num_gpus=2, torch_dtype=torch.float16
 )
 model = SparseAutoModelForCausalLM.from_pretrained(
-    model_stub, torch_dtype=torch.float16, device_map=device_map
 )
 output_dir = f"./{model_name}-FP8-dynamic"
@@ -140,7 +140,7 @@ oneshot(
 The model was evaluated on MMLU, ARC-Challenge, GSM-8K, Hellaswag, Winogrande and TruthfulQA.
 Evaluation was conducted using the Neural Magic fork of [lm-evaluation-harness](https://github.com/neuralmagic/lm-evaluation-harness/tree/llama_3.1_instruct) (branch llama_3.1_instruct) and the [vLLM](https://docs.vllm.ai/en/stable/) engine.
-This version of the lm-evaluation-harness includes versions of ARC-Challenge and GSM-8K that match the prompting style of [Meta-Llama-3.1-Instruct-evals](https://huggingface.co/datasets/meta-llama/Meta-Llama-3.1-8B-Instruct-evals).
 ### Accuracy
@@ -151,7 +151,7 @@ This version of the lm-evaluation-harness includes versions of ARC-Challenge and
    </td>
    <td><strong>Meta-Llama-3.1-70B-Instruct </strong>
    </td>
-   <td><strong>Meta-Llama-3.1-70B-Instruct-FP8-dynamic(this model)</strong>
    </td>
    <td><strong>Recovery</strong>
    </td>
@@ -159,71 +159,81 @@ This version of the lm-evaluation-harness includes versions of ARC-Challenge and
   <tr>
    <td>MMLU (5-shot)
    </td>
-   <td>82.21
    </td>
-   <td>82.13
    </td>
    <td>99.90%
    </td>
   </tr>
   <tr>
    <td>ARC Challenge (0-shot)
    </td>
-   <td>95.05
    </td>
-   <td>94.88
    </td>
-   <td>99.82%
    </td>
   </tr>
   <tr>
-   <td>GSM-8K (CoT, 8-shot, strict-match)
    </td>
-   <td>93.18
    </td>
-   <td>92.19
    </td>
-   <td>98.94%
    </td>
   </tr>
   <tr>
    <td>Hellaswag (10-shot)
    </td>
-   <td>86.33
    </td>
-   <td>86.27
    </td>
-   <td>99.93%
    </td>
   </tr>
   <tr>
    <td>Winogrande (5-shot)
    </td>
-   <td>85.00
    </td>
-   <td>85.00
    </td>
-   <td>100.0%
    </td>
   </tr>
   <tr>
    <td>TruthfulQA (0-shot, mc2)
    </td>
-   <td>59.90
    </td>
-   <td>60.01
    </td>
-   <td>100.1%
    </td>
   </tr>
   <tr>
    <td><strong>Average</strong>
    </td>
-   <td><strong>83.61</strong>
    </td>
-   <td><strong>83.41</strong>
    </td>
-   <td><strong>99.76%</strong>
    </td>
   </tr>
 </table>
@@ -242,6 +252,17 @@ lm_eval \
   --batch_size auto
 ```
 #### ARC-Challenge
 ```
 lm_eval \

 - **Model Developers:** Neural Magic
 Quantized version of [Meta-Llama-3.1-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-70B-Instruct).
+It achieves an average score of 84.16 on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) benchmark (version 1), whereas the unquantized model achieves 84.40.
 ### Model Optimizations
 model_name = model_stub.split("/")[-1]
 device_map = calculate_offload_device_map(
+    model_stub, reserve_for_hessians=False, num_gpus=2, torch_dtype="auto"
 )
 model = SparseAutoModelForCausalLM.from_pretrained(
+    model_stub, torch_dtype="auto", device_map=device_map
 )
 output_dir = f"./{model_name}-FP8-dynamic"
 The model was evaluated on MMLU, ARC-Challenge, GSM-8K, Hellaswag, Winogrande and TruthfulQA.
 Evaluation was conducted using the Neural Magic fork of [lm-evaluation-harness](https://github.com/neuralmagic/lm-evaluation-harness/tree/llama_3.1_instruct) (branch llama_3.1_instruct) and the [vLLM](https://docs.vllm.ai/en/stable/) engine.
+This version of the lm-evaluation-harness includes versions of ARC-Challenge, GSM-8K, MMLU, and MMLU-cot that match the prompting style of [Meta-Llama-3.1-Instruct-evals](https://huggingface.co/datasets/meta-llama/Meta-Llama-3.1-8B-Instruct-evals).
 ### Accuracy
    </td>
    <td><strong>Meta-Llama-3.1-70B-Instruct </strong>
    </td>
+   <td><strong>Meta-Llama-3.1-70B-Instruct-FP8(this model)</strong>
    </td>
    <td><strong>Recovery</strong>
    </td>
   <tr>
    <td>MMLU (5-shot)
    </td>
+   <td>83.83
    </td>
+   <td>83.75
    </td>
    <td>99.90%
    </td>
   </tr>
+  <tr>
+   <td>MMLU-cot (0-shot)
+   </td>
+   <td>86.01
+   </td>
+   <td>85.48
+   </td>
+   <td>99.38%
+   </td>
+  </tr>
   <tr>
    <td>ARC Challenge (0-shot)
    </td>
+   <td>93.26
    </td>
+   <td>93.52
    </td>
+   <td>100.2%
    </td>
   </tr>
   <tr>
+   <td>GSM-8K-cot (8-shot, strict-match)
    </td>
+   <td>94.92
    </td>
+   <td>94.54
    </td>
+   <td>99.60%
    </td>
   </tr>
   <tr>
    <td>Hellaswag (10-shot)
    </td>
+   <td>86.75
    </td>
+   <td>86.63
    </td>
+   <td>99.86%
    </td>
   </tr>
   <tr>
    <td>Winogrande (5-shot)
    </td>
+   <td>85.32
    </td>
+   <td>84.61
    </td>
+   <td>99.17%
    </td>
   </tr>
   <tr>
    <td>TruthfulQA (0-shot, mc2)
    </td>
+   <td>60.68
    </td>
+   <td>60.60
    </td>
+   <td>99.87%
    </td>
   </tr>
   <tr>
    <td><strong>Average</strong>
    </td>
+   <td><strong>84.40</strong>
    </td>
+   <td><strong>84.16</strong>
    </td>
+   <td><strong>99.72%</strong>
    </td>
   </tr>
 </table>
   --batch_size auto
 ```
+#### MMLU-cot
+```
+lm_eval \
+  --model vllm \
+  --model_args pretrained="neuralmagic/Meta-Llama-3.1-70B-Instruct-FP8-dynamic",dtype=auto,add_bos_token=True,max_model_len=4096,tensor_parallel_size=2 \
+  --tasks mmlu_cot_0shot_llama_3.1_instruct \
+  --apply_chat_template \
+  --num_fewshot 0 \
+  --batch_size auto
+```
 #### ARC-Challenge
 ```
 lm_eval \