Instructions to use chargoddard/Meta-Llama-3-8B-InitializedEmbeds with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use chargoddard/Meta-Llama-3-8B-InitializedEmbeds with Transformers:

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

pipe = pipeline("text-generation", model="chargoddard/Meta-Llama-3-8B-InitializedEmbeds")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("chargoddard/Meta-Llama-3-8B-InitializedEmbeds")
model = AutoModelForCausalLM.from_pretrained("chargoddard/Meta-Llama-3-8B-InitializedEmbeds")
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]:]))

Inference
Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use chargoddard/Meta-Llama-3-8B-InitializedEmbeds with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "chargoddard/Meta-Llama-3-8B-InitializedEmbeds"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "chargoddard/Meta-Llama-3-8B-InitializedEmbeds",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/chargoddard/Meta-Llama-3-8B-InitializedEmbeds

SGLang

How to use chargoddard/Meta-Llama-3-8B-InitializedEmbeds 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 "chargoddard/Meta-Llama-3-8B-InitializedEmbeds" \
    --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": "chargoddard/Meta-Llama-3-8B-InitializedEmbeds",
		"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 "chargoddard/Meta-Llama-3-8B-InitializedEmbeds" \
        --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": "chargoddard/Meta-Llama-3-8B-InitializedEmbeds",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use chargoddard/Meta-Llama-3-8B-InitializedEmbeds with Docker Model Runner:
```
docker model run hf.co/chargoddard/Meta-Llama-3-8B-InitializedEmbeds
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Meta-Llama-3-8B-InitializedEmbeds

This is just Llama-3-8B with the embeddings for special tokens copied from the Instruct version. Should behave pretty much identically to the base model, but with less glossolalia when it encounters <|start_header_id|> and the like.

I'm using this as a base to fine tune. Having these embeddings reasonable instead of randomly initialized should give a smoother start.

This is a merge of pre-trained language models created using mergekit.

Merge Details

Merge Method

This model was merged using the linear merge method.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

merge_method: linear
dtype: float32
out_dtype: bfloat16
models:
  - model: NousResearch/Meta-Llama-3-8B
    parameters:
      weight: 1.0
  - model: NousResearch/Meta-Llama-3-8B-Instruct
    parameters:
      weight: 0.0
tokenizer:
  source: NousResearch/Meta-Llama-3-8B-Instruct
  tokens:
    <|start_header_id|>:
      source: NousResearch/Meta-Llama-3-8B-Instruct
      force: true
    <|end_header_id|>:
      source: NousResearch/Meta-Llama-3-8B-Instruct
      force: true
    <|eot_id|>:
      source: NousResearch/Meta-Llama-3-8B-Instruct
      force: true
    <|end_of_text|>:
      source: NousResearch/Meta-Llama-3-8B-Instruct
      force: true
    <|begin_of_text|>:
      source: NousResearch/Meta-Llama-3-8B-Instruct
      force: true

Downloads last month: 10

Safetensors

Model size

8B params

Tensor type

BF16

Model tree for chargoddard/Meta-Llama-3-8B-InitializedEmbeds

NousResearch/Meta-Llama-3-8B

NousResearch/Meta-Llama-3-8B-Instruct

Merge model

this model

Quantizations

1 model

Paper for chargoddard/Meta-Llama-3-8B-InitializedEmbeds

Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time

Paper • 2203.05482 • Published Mar 10, 2022 • 8