Instructions to use tsavage68/UTI_L3_1000steps_1e8rate_03beta_CSFTDPO with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tsavage68/UTI_L3_1000steps_1e8rate_03beta_CSFTDPO with Transformers:

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

pipe = pipeline("text-generation", model="tsavage68/UTI_L3_1000steps_1e8rate_03beta_CSFTDPO")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("tsavage68/UTI_L3_1000steps_1e8rate_03beta_CSFTDPO")
model = AutoModelForCausalLM.from_pretrained("tsavage68/UTI_L3_1000steps_1e8rate_03beta_CSFTDPO", device_map="auto")
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 tsavage68/UTI_L3_1000steps_1e8rate_03beta_CSFTDPO with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tsavage68/UTI_L3_1000steps_1e8rate_03beta_CSFTDPO"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tsavage68/UTI_L3_1000steps_1e8rate_03beta_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tsavage68/UTI_L3_1000steps_1e8rate_03beta_CSFTDPO

SGLang

How to use tsavage68/UTI_L3_1000steps_1e8rate_03beta_CSFTDPO 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 "tsavage68/UTI_L3_1000steps_1e8rate_03beta_CSFTDPO" \
    --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": "tsavage68/UTI_L3_1000steps_1e8rate_03beta_CSFTDPO",
		"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 "tsavage68/UTI_L3_1000steps_1e8rate_03beta_CSFTDPO" \
        --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": "tsavage68/UTI_L3_1000steps_1e8rate_03beta_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use tsavage68/UTI_L3_1000steps_1e8rate_03beta_CSFTDPO with Docker Model Runner:
```
docker model run hf.co/tsavage68/UTI_L3_1000steps_1e8rate_03beta_CSFTDPO
```

UTI_L3_1000steps_1e8rate_03beta_CSFTDPO

This model is a fine-tuned version of tsavage68/UTI_L3_1000steps_1e5rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 0.6892
Rewards/chosen: 0.0019
Rewards/rejected: -0.0064
Rewards/accuracies: 0.6200
Rewards/margins: 0.0083
Logps/rejected: -63.2161
Logps/chosen: -32.4727
Logits/rejected: -1.3229
Logits/chosen: -1.3077

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: 1e-08
train_batch_size: 2
eval_batch_size: 1
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 4
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 100
training_steps: 1000

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6937	0.3333	25	0.6946	-0.0008	0.0021	0.0600	-0.0029	-63.1876	-32.4816	-1.3229	-1.3077
0.6914	0.6667	50	0.6957	-0.0038	0.0008	0.4400	-0.0046	-63.1920	-32.4917	-1.3228	-1.3077
0.691	1.0	75	0.6939	-0.0059	-0.0048	0.4400	-0.0011	-63.2107	-32.4987	-1.3231	-1.3080
0.6895	1.3333	100	0.6936	-0.0030	-0.0027	0.4600	-0.0004	-63.2035	-32.4892	-1.3230	-1.3079
0.6875	1.6667	125	0.6931	0.0025	0.0020	0.5100	0.0006	-63.1881	-32.4706	-1.3230	-1.3079
0.6949	2.0	150	0.6956	0.0004	0.0046	0.4400	-0.0042	-63.1792	-32.4777	-1.3229	-1.3078
0.6996	2.3333	175	0.6922	-0.0011	-0.0034	0.5	0.0023	-63.2060	-32.4828	-1.3229	-1.3078
0.691	2.6667	200	0.6933	0.0001	0.0001	0.5200	-0.0000	-63.1942	-32.4786	-1.3230	-1.3079
0.6879	3.0	225	0.6925	-0.0011	-0.0031	0.5400	0.0020	-63.2049	-32.4826	-1.3230	-1.3079
0.691	3.3333	250	0.6907	0.0015	-0.0040	0.4900	0.0055	-63.2080	-32.4741	-1.3229	-1.3079
0.6953	3.6667	275	0.6924	0.0027	0.0008	0.4700	0.0019	-63.1921	-32.4699	-1.3229	-1.3078
0.6906	4.0	300	0.6906	-0.0010	-0.0066	0.5200	0.0056	-63.2167	-32.4825	-1.3230	-1.3079
0.6973	4.3333	325	0.6879	0.0027	-0.0083	0.6100	0.0111	-63.2224	-32.4699	-1.3229	-1.3078
0.6887	4.6667	350	0.6875	0.0051	-0.0066	0.5900	0.0118	-63.2168	-32.4619	-1.3230	-1.3078
0.6891	5.0	375	0.6887	0.0018	-0.0076	0.5800	0.0093	-63.2199	-32.4732	-1.3228	-1.3077
0.6961	5.3333	400	0.6906	0.0023	-0.0033	0.5700	0.0055	-63.2056	-32.4714	-1.3230	-1.3079
0.6848	5.6667	425	0.6902	0.0003	-0.0061	0.5200	0.0064	-63.2151	-32.4779	-1.3229	-1.3078
0.6855	6.0	450	0.6883	0.0021	-0.0083	0.5600	0.0104	-63.2224	-32.4722	-1.3230	-1.3079
0.6898	6.3333	475	0.6922	-0.0013	-0.0038	0.5300	0.0026	-63.2075	-32.4832	-1.3229	-1.3078
0.6887	6.6667	500	0.6905	0.0023	-0.0037	0.5400	0.0060	-63.2071	-32.4715	-1.3229	-1.3078
0.6918	7.0	525	0.6862	0.0033	-0.0110	0.5900	0.0144	-63.2315	-32.4679	-1.3231	-1.3080
0.6871	7.3333	550	0.6902	0.0020	-0.0043	0.5300	0.0063	-63.2090	-32.4723	-1.3229	-1.3078
0.6879	7.6667	575	0.6927	-0.0028	-0.0041	0.4800	0.0013	-63.2085	-32.4885	-1.3229	-1.3078
0.6793	8.0	600	0.6925	-0.0004	-0.0022	0.4600	0.0018	-63.2021	-32.4805	-1.3230	-1.3079
0.6918	8.3333	625	0.6904	0.0009	-0.0052	0.5200	0.0060	-63.2119	-32.4762	-1.3230	-1.3079
0.6887	8.6667	650	0.6896	0.0015	-0.0061	0.5500	0.0076	-63.2150	-32.4739	-1.3229	-1.3078
0.6965	9.0	675	0.6905	-0.0013	-0.0072	0.5600	0.0060	-63.2188	-32.4833	-1.3230	-1.3078
0.6895	9.3333	700	0.6877	0.0038	-0.0076	0.6200	0.0114	-63.2200	-32.4662	-1.3229	-1.3078
0.6855	9.6667	725	0.6891	0.0014	-0.0074	0.5500	0.0087	-63.2192	-32.4744	-1.3229	-1.3078
0.6871	10.0	750	0.6879	0.0033	-0.0077	0.5900	0.0110	-63.2204	-32.4679	-1.3230	-1.3078
0.6887	10.3333	775	0.6881	0.0034	-0.0072	0.6200	0.0106	-63.2186	-32.4675	-1.3229	-1.3077
0.693	10.6667	800	0.6890	0.0023	-0.0065	0.6200	0.0088	-63.2163	-32.4715	-1.3229	-1.3078
0.6875	11.0	825	0.6892	0.0019	-0.0064	0.6200	0.0083	-63.2161	-32.4727	-1.3229	-1.3077
0.6895	11.3333	850	0.6892	0.0019	-0.0064	0.6200	0.0083	-63.2161	-32.4727	-1.3229	-1.3077
0.6887	11.6667	875	0.6892	0.0019	-0.0064	0.6200	0.0083	-63.2161	-32.4727	-1.3229	-1.3077
0.6918	12.0	900	0.6892	0.0019	-0.0064	0.6200	0.0083	-63.2161	-32.4727	-1.3229	-1.3077
0.6918	12.3333	925	0.6892	0.0019	-0.0064	0.6200	0.0083	-63.2161	-32.4727	-1.3229	-1.3077
0.6816	12.6667	950	0.6892	0.0019	-0.0064	0.6200	0.0083	-63.2161	-32.4727	-1.3229	-1.3077
0.6883	13.0	975	0.6892	0.0019	-0.0064	0.6200	0.0083	-63.2161	-32.4727	-1.3229	-1.3077
0.6883	13.3333	1000	0.6892	0.0019	-0.0064	0.6200	0.0083	-63.2161	-32.4727	-1.3229	-1.3077

Framework versions

Transformers 4.41.2
Pytorch 2.0.0+cu117
Datasets 2.19.2
Tokenizers 0.19.1

Downloads last month: 6

Safetensors

Model size

8B params

Tensor type

F16

Model tree for tsavage68/UTI_L3_1000steps_1e8rate_03beta_CSFTDPO

Base model

meta-llama/Meta-Llama-3-8B-Instruct

Finetuned

tsavage68/UTI_L3_1000steps_1e5rate_SFT

Finetuned

(39)

this model