Instructions to use phatvo/Meta-Llama3.1-8B-Instruct-RAFT with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use phatvo/Meta-Llama3.1-8B-Instruct-RAFT with Transformers:

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

pipe = pipeline("text-generation", model="phatvo/Meta-Llama3.1-8B-Instruct-RAFT")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("phatvo/Meta-Llama3.1-8B-Instruct-RAFT")
model = AutoModelForCausalLM.from_pretrained("phatvo/Meta-Llama3.1-8B-Instruct-RAFT")
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 Settings

vLLM

How to use phatvo/Meta-Llama3.1-8B-Instruct-RAFT with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "phatvo/Meta-Llama3.1-8B-Instruct-RAFT"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "phatvo/Meta-Llama3.1-8B-Instruct-RAFT",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/phatvo/Meta-Llama3.1-8B-Instruct-RAFT

SGLang

How to use phatvo/Meta-Llama3.1-8B-Instruct-RAFT 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 "phatvo/Meta-Llama3.1-8B-Instruct-RAFT" \
    --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": "phatvo/Meta-Llama3.1-8B-Instruct-RAFT",
		"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 "phatvo/Meta-Llama3.1-8B-Instruct-RAFT" \
        --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": "phatvo/Meta-Llama3.1-8B-Instruct-RAFT",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use phatvo/Meta-Llama3.1-8B-Instruct-RAFT with Docker Model Runner:
```
docker model run hf.co/phatvo/Meta-Llama3.1-8B-Instruct-RAFT
```

Model Card for Model ID

Model Details

Model Description

LORA adapters of meta-llama/Meta-Llama-3.1-8B-Instruct, trained on 100 context samples from the HotpotQA dataset using the RAFT method, enable the model to better reason through the context and return more accurate outcomes.

Evaluation

Evaluated on FULL validation set of HotpotQA.

type	exatch_match	f1	precision	recall
pretrained	0.2980	0.3979	0.4116	0.5263
finetuned	0.3606	0.4857	0.4989	0.5318

Finetuned version increases 22% on F1 and 15% on average

Usage

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline

model_id = "phatvo/Meta-Llama3.1-8B-Instruct-RAFT"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
  model_id, device_map="auto", revision="main", trust_remote_code=True)
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)

inst = "Given the question and context below, thinking in logical reasoning way for your answer.\
Please provide only your answer in this format: CoT Answer: {reason} <ANSWER>: {answer}."
context = ""
question = ""
prompt = f"{context}\n{question}"

chat = [
        {"role": "system", "content": inst},
        {"role": "user", "content": prompt},
    ]
prompt = tokenizer.apply_chat_template(chat, tokenize=False)

output = pipe(prompt,
  temperature=0.001, 
  max_new_tokens=1024, # recommended to set it more than 800 
  return_full_text=False,
  do_sample=True)

print(output[0]["generated_text"])
# CoT Answer: thoughts... <ANSWER>: final_answer...

Downloads last month: 2

Safetensors

Model size

8B params

Tensor type

BF16

Dataset used to train phatvo/Meta-Llama3.1-8B-Instruct-RAFT

Collection including phatvo/Meta-Llama3.1-8B-Instruct-RAFT

RAFT

Collection

RAFT finetuned models • 4 items • Updated Aug 21, 2024