Instructions to use tsavage68/UTI_L3_1000steps_1e8rate_01beta_CSFTDPO with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tsavage68/UTI_L3_1000steps_1e8rate_01beta_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_01beta_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_01beta_CSFTDPO")
model = AutoModelForCausalLM.from_pretrained("tsavage68/UTI_L3_1000steps_1e8rate_01beta_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_01beta_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_01beta_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_01beta_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tsavage68/UTI_L3_1000steps_1e8rate_01beta_CSFTDPO

SGLang

How to use tsavage68/UTI_L3_1000steps_1e8rate_01beta_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_01beta_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_01beta_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_01beta_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_01beta_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use tsavage68/UTI_L3_1000steps_1e8rate_01beta_CSFTDPO with Docker Model Runner:
```
docker model run hf.co/tsavage68/UTI_L3_1000steps_1e8rate_01beta_CSFTDPO
```

UTI_L3_1000steps_1e8rate_01beta_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: 0.6922
Rewards/chosen: -0.0001
Rewards/rejected: -0.0021
Rewards/accuracies: 0.5700
Rewards/margins: 0.0020
Logps/rejected: -63.2152
Logps/chosen: -32.4800
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	0.6936	-0.0003	0.0007	0.0500	-0.0010	-63.1876	-32.4816	-1.3228	-1.3077
0.6934	0.6667	50	0.6924	-0.0003	-0.0019	0.5800	0.0016	-63.2137	-32.4817	-1.3229	-1.3078
0.6945	1.0	75	0.6927	-0.0018	-0.0026	0.4600	0.0008	-63.2208	-32.4966	-1.3232	-1.3080
0.6926	1.3333	100	0.6933	-0.0002	0.0001	0.5	-0.0003	-63.1934	-32.4807	-1.3230	-1.3079
0.6918	1.6667	125	0.6926	-0.0001	-0.0013	0.5200	0.0012	-63.2078	-32.4798	-1.3229	-1.3078
0.6922	2.0	150	0.6927	-0.0004	-0.0013	0.5	0.0010	-63.2080	-32.4826	-1.3231	-1.3080
0.6941	2.3333	175	0.6927	-0.0003	-0.0011	0.5200	0.0009	-63.2060	-32.4817	-1.3229	-1.3077
0.6926	2.6667	200	0.6934	-0.0015	-0.0010	0.4300	-0.0005	-63.2044	-32.4940	-1.3230	-1.3079
0.6922	3.0	225	0.6928	0.0006	-0.0002	0.5500	0.0008	-63.1962	-32.4728	-1.3231	-1.3079
0.6918	3.3333	250	0.6922	0.0011	-0.0008	0.5700	0.0020	-63.2029	-32.4677	-1.3230	-1.3079
0.6926	3.6667	275	0.6926	0.0004	-0.0008	0.4900	0.0011	-63.2022	-32.4752	-1.3229	-1.3078
0.6906	4.0	300	0.6923	0.0000	-0.0017	0.4600	0.0017	-63.2119	-32.4789	-1.3231	-1.3079
0.6934	4.3333	325	0.6926	-0.0006	-0.0018	0.5	0.0012	-63.2131	-32.4852	-1.3231	-1.3079
0.6918	4.6667	350	0.6921	0.0014	-0.0008	0.5200	0.0022	-63.2022	-32.4648	-1.3231	-1.3080
0.6918	5.0	375	0.6917	-0.0002	-0.0033	0.5600	0.0030	-63.2273	-32.4813	-1.3230	-1.3079
0.6922	5.3333	400	0.6930	-0.0006	-0.0009	0.4800	0.0003	-63.2034	-32.4851	-1.3231	-1.3080
0.693	5.6667	425	0.6923	0.0005	-0.0013	0.5200	0.0018	-63.2075	-32.4743	-1.3230	-1.3080
0.6906	6.0	450	0.6916	0.0007	-0.0024	0.5900	0.0031	-63.2182	-32.4716	-1.3231	-1.3080
0.6898	6.3333	475	0.6915	0.0002	-0.0033	0.5700	0.0034	-63.2273	-32.4774	-1.3228	-1.3078
0.6922	6.6667	500	0.6925	0.0003	-0.0012	0.5400	0.0014	-63.2066	-32.4765	-1.3230	-1.3079
0.6918	7.0	525	0.6915	0.0006	-0.0027	0.4900	0.0033	-63.2220	-32.4735	-1.3231	-1.3079
0.6914	7.3333	550	0.6922	0.0005	-0.0015	0.5300	0.0020	-63.2102	-32.4742	-1.3229	-1.3079
0.6906	7.6667	575	0.6919	0.0002	-0.0024	0.5400	0.0026	-63.2189	-32.4772	-1.3230	-1.3079
0.691	8.0	600	0.6930	-0.0006	-0.0010	0.5400	0.0004	-63.2047	-32.4854	-1.3229	-1.3078
0.6922	8.3333	625	0.6918	0.0001	-0.0027	0.5600	0.0028	-63.2220	-32.4781	-1.3230	-1.3079
0.6918	8.6667	650	0.6921	0.0012	-0.0009	0.5200	0.0021	-63.2039	-32.4669	-1.3230	-1.3078
0.6922	9.0	675	0.6922	0.0012	-0.0007	0.6100	0.0020	-63.2019	-32.4667	-1.3230	-1.3079
0.6934	9.3333	700	0.6920	-0.0001	-0.0025	0.5100	0.0024	-63.2195	-32.4799	-1.3230	-1.3079
0.6895	9.6667	725	0.6926	0.0005	-0.0007	0.5	0.0012	-63.2018	-32.4743	-1.3230	-1.3080
0.6918	10.0	750	0.6919	0.0004	-0.0022	0.5600	0.0025	-63.2163	-32.4752	-1.3230	-1.3078
0.6914	10.3333	775	0.6920	-0.0000	-0.0023	0.5300	0.0023	-63.2175	-32.4793	-1.3229	-1.3078
0.6934	10.6667	800	0.6920	0.0001	-0.0022	0.5600	0.0023	-63.2163	-32.4776	-1.3229	-1.3078
0.6926	11.0	825	0.6922	-0.0001	-0.0021	0.5700	0.0020	-63.2152	-32.4800	-1.3229	-1.3078
0.6934	11.3333	850	0.6922	-0.0001	-0.0021	0.5700	0.0020	-63.2152	-32.4800	-1.3229	-1.3078
0.6914	11.6667	875	0.6922	-0.0001	-0.0021	0.5700	0.0020	-63.2152	-32.4800	-1.3229	-1.3078
0.6918	12.0	900	0.6922	-0.0001	-0.0021	0.5700	0.0020	-63.2152	-32.4800	-1.3229	-1.3078
0.6891	12.3333	925	0.6922	-0.0001	-0.0021	0.5700	0.0020	-63.2152	-32.4800	-1.3229	-1.3078
0.6918	12.6667	950	0.6922	-0.0001	-0.0021	0.5700	0.0020	-63.2152	-32.4800	-1.3229	-1.3078
0.691	13.0	975	0.6922	-0.0001	-0.0021	0.5700	0.0020	-63.2152	-32.4800	-1.3229	-1.3078
0.6902	13.3333	1000	0.6922	-0.0001	-0.0021	0.5700	0.0020	-63.2152	-32.4800	-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: 6

Safetensors

Model size

8B params

Tensor type

F16

Model tree for tsavage68/UTI_L3_1000steps_1e8rate_01beta_CSFTDPO

Base model

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

Finetuned

tsavage68/UTI_L3_1000steps_1e5rate_SFT

Finetuned

(39)

this model