Instructions to use tsavage68/UTI_L3_1000steps_1e8rate_05beta_CSFTDPO with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tsavage68/UTI_L3_1000steps_1e8rate_05beta_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_05beta_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_05beta_CSFTDPO")
model = AutoModelForCausalLM.from_pretrained("tsavage68/UTI_L3_1000steps_1e8rate_05beta_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_05beta_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_05beta_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_05beta_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tsavage68/UTI_L3_1000steps_1e8rate_05beta_CSFTDPO

SGLang

How to use tsavage68/UTI_L3_1000steps_1e8rate_05beta_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_05beta_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_05beta_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_05beta_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_05beta_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use tsavage68/UTI_L3_1000steps_1e8rate_05beta_CSFTDPO with Docker Model Runner:
```
docker model run hf.co/tsavage68/UTI_L3_1000steps_1e8rate_05beta_CSFTDPO
```

UTI_L3_1000steps_1e8rate_05beta_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: 0.6943
Rewards/chosen: -0.0022
Rewards/rejected: -0.0014
Rewards/accuracies: 0.4700
Rewards/margins: -0.0008
Logps/rejected: -63.1976
Logps/chosen: -32.4834
Logits/rejected: -1.3229
Logits/chosen: -1.3078

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	0.6957	-0.0013	0.0035	0.0500	-0.0048	-63.1876	-32.4816	-1.3228	-1.3077
0.6934	0.6667	50	0.6925	-0.0026	-0.0053	0.5100	0.0027	-63.2053	-32.4843	-1.3228	-1.3077
0.6898	1.0	75	0.6961	-0.0017	0.0030	0.4800	-0.0047	-63.1886	-32.4823	-1.3230	-1.3078
0.6855	1.3333	100	0.6937	0.0021	0.0020	0.5200	0.0001	-63.1908	-32.4748	-1.3230	-1.3079
0.6852	1.6667	125	0.6971	-0.0047	0.0016	0.4200	-0.0062	-63.1916	-32.4884	-1.3230	-1.3078
0.6879	2.0	150	0.6905	0.0038	-0.0031	0.4500	0.0069	-63.2009	-32.4715	-1.3231	-1.3079
0.6957	2.3333	175	0.6911	-0.0001	-0.0058	0.5200	0.0057	-63.2062	-32.4793	-1.3230	-1.3079
0.6988	2.6667	200	0.6929	-0.0035	-0.0053	0.5300	0.0018	-63.2052	-32.4859	-1.3229	-1.3078
0.6926	3.0	225	0.6903	0.0001	-0.0071	0.5	0.0072	-63.2088	-32.4787	-1.3230	-1.3078
0.6895	3.3333	250	0.6896	-0.0014	-0.0101	0.5	0.0087	-63.2149	-32.4818	-1.3230	-1.3079
0.6988	3.6667	275	0.6932	-0.0029	-0.0044	0.5	0.0015	-63.2034	-32.4847	-1.3231	-1.3079
0.6719	4.0	300	0.6895	-0.0022	-0.0107	0.4900	0.0085	-63.2161	-32.4833	-1.3230	-1.3079
0.6988	4.3333	325	0.6886	0.0061	-0.0045	0.5100	0.0106	-63.2038	-32.4668	-1.3231	-1.3080
0.6859	4.6667	350	0.6869	0.0014	-0.0126	0.5500	0.0139	-63.2198	-32.4762	-1.3231	-1.3080
0.6922	5.0	375	0.6888	-0.0004	-0.0102	0.5	0.0097	-63.2150	-32.4799	-1.3230	-1.3079
0.6937	5.3333	400	0.6875	0.0028	-0.0102	0.5400	0.0130	-63.2150	-32.4734	-1.3231	-1.3080
0.6773	5.6667	425	0.6857	0.0025	-0.0143	0.5300	0.0168	-63.2233	-32.4741	-1.3228	-1.3078
0.684	6.0	450	0.6900	0.0039	-0.0036	0.5400	0.0075	-63.2019	-32.4713	-1.3231	-1.3079
0.6914	6.3333	475	0.6902	0.0001	-0.0078	0.5300	0.0079	-63.2103	-32.4789	-1.3230	-1.3079
0.6879	6.6667	500	0.6871	0.0049	-0.0084	0.5300	0.0133	-63.2115	-32.4691	-1.3229	-1.3078
0.6934	7.0	525	0.6896	0.0039	-0.0046	0.4900	0.0085	-63.2039	-32.4712	-1.3230	-1.3079
0.6887	7.3333	550	0.6901	0.0037	-0.0042	0.5200	0.0079	-63.2031	-32.4717	-1.3230	-1.3079
0.6863	7.6667	575	0.6909	-0.0015	-0.0071	0.5800	0.0057	-63.2090	-32.4819	-1.3230	-1.3079
0.6809	8.0	600	0.6895	-0.0005	-0.0093	0.5500	0.0088	-63.2133	-32.4801	-1.3229	-1.3077
0.6879	8.3333	625	0.6906	0.0042	-0.0019	0.5200	0.0061	-63.1984	-32.4706	-1.3230	-1.3079
0.6844	8.6667	650	0.6865	-0.0004	-0.0156	0.5100	0.0152	-63.2259	-32.4798	-1.3229	-1.3079
0.6945	9.0	675	0.6899	-0.0047	-0.0124	0.5500	0.0077	-63.2195	-32.4884	-1.3230	-1.3079
0.6918	9.3333	700	0.6859	0.0034	-0.0127	0.5400	0.0160	-63.2200	-32.4723	-1.3230	-1.3079
0.6848	9.6667	725	0.6909	-0.0053	-0.0113	0.5200	0.0060	-63.2172	-32.4896	-1.3229	-1.3078
0.6801	10.0	750	0.6915	0.0025	-0.0025	0.5300	0.0049	-63.1997	-32.4741	-1.3229	-1.3078
0.684	10.3333	775	0.6939	-0.0003	-0.0002	0.4900	-0.0001	-63.1951	-32.4797	-1.3229	-1.3078
0.6891	10.6667	800	0.6936	-0.0012	-0.0017	0.4900	0.0005	-63.1981	-32.4814	-1.3229	-1.3078
0.6883	11.0	825	0.6943	-0.0022	-0.0014	0.4700	-0.0008	-63.1976	-32.4834	-1.3229	-1.3078
0.6969	11.3333	850	0.6943	-0.0022	-0.0014	0.4700	-0.0008	-63.1976	-32.4834	-1.3229	-1.3078
0.6984	11.6667	875	0.6943	-0.0022	-0.0014	0.4700	-0.0008	-63.1976	-32.4834	-1.3229	-1.3078
0.6937	12.0	900	0.6943	-0.0022	-0.0014	0.4700	-0.0008	-63.1976	-32.4834	-1.3229	-1.3078
0.684	12.3333	925	0.6943	-0.0022	-0.0014	0.4700	-0.0008	-63.1976	-32.4834	-1.3229	-1.3078
0.682	12.6667	950	0.6943	-0.0022	-0.0014	0.4700	-0.0008	-63.1976	-32.4834	-1.3229	-1.3078
0.6863	13.0	975	0.6943	-0.0022	-0.0014	0.4700	-0.0008	-63.1976	-32.4834	-1.3229	-1.3078
0.6836	13.3333	1000	0.6943	-0.0022	-0.0014	0.4700	-0.0008	-63.1976	-32.4834	-1.3229	-1.3078

Framework versions

Transformers 4.41.2
Pytorch 2.0.0+cu117
Datasets 2.19.2
Tokenizers 0.19.1

Downloads last month: 9

Safetensors

Model size

8B params

Tensor type

F16

Model tree for tsavage68/UTI_L3_1000steps_1e8rate_05beta_CSFTDPO

Base model

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

Finetuned

tsavage68/UTI_L3_1000steps_1e5rate_SFT

Finetuned

(39)

this model