Instructions to use Trelis/99-v9 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Trelis/99-v9 with Transformers:

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

pipe = pipeline("text-generation", model="Trelis/99-v9")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Trelis/99-v9")
model = AutoModelForCausalLM.from_pretrained("Trelis/99-v9")
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 Trelis/99-v9 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Trelis/99-v9"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Trelis/99-v9",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Trelis/99-v9

SGLang

How to use Trelis/99-v9 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 "Trelis/99-v9" \
    --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": "Trelis/99-v9",
		"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 "Trelis/99-v9" \
        --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": "Trelis/99-v9",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Trelis/99-v9 with Docker Model Runner:
```
docker model run hf.co/Trelis/99-v9
```

99-v9

This model is a fine-tuned version of Trelis/SmolLM-135M-Instruct-layer-pruned-90M-raw on the None dataset. It achieves the following results on the evaluation set:

Loss: 0.7495

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.002
train_batch_size: 8
eval_batch_size: 8
seed: 42
distributed_type: multi-GPU
num_devices: 8
gradient_accumulation_steps: 4
total_train_batch_size: 256
total_eval_batch_size: 64
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.005
lr_scheduler_warmup_steps: 89
training_steps: 17894

Training results

Training Loss	Epoch	Step	Validation Loss
0.6331	0.0500	894	0.6004
0.5667	0.0999	1788	0.5463
0.5423	0.1499	2682	0.5138
0.5749	0.1998	3576	0.7377
0.5378	0.2498	4470	0.7542
0.506	0.2998	5364	0.7902
0.5561	0.3497	6258	0.7810
0.5259	0.3997	7152	0.7914
0.5516	0.4496	8046	0.7611
0.5131	0.4996	8940	0.6860
0.5069	0.5496	9834	0.7247
0.4977	0.5995	10728	0.7375
0.4976	0.6495	11622	0.7436
0.5018	0.6995	12516	0.7520
0.537	0.7494	13410	0.7613
0.5018	0.7994	14304	0.6922
0.4891	0.8493	15198	0.7322
0.4808	0.8993	16092	0.7430
0.5231	0.9493	16986	0.7546
0.5103	0.9992	17880	0.7495

Framework versions

Transformers 4.44.2
Pytorch 2.1.1+cu121
Datasets 3.0.0
Tokenizers 0.19.1

Downloads last month: 1

Safetensors

Model size

99.1M params

Tensor type

BF16

Model tree for Trelis/99-v9

Base model

HuggingFaceTB/SmolLM-135M

Quantized

HuggingFaceTB/SmolLM-135M-Instruct

Finetuned

Trelis/SmolLM-135M-Instruct-layer-pruned-90M-raw

Finetuned

(2)

this model