Instructions to use tsavage68/Summary_L3_1000steps_1e7rate_01beta_CSFTDPO with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tsavage68/Summary_L3_1000steps_1e7rate_01beta_CSFTDPO with Transformers:

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

pipe = pipeline("text-generation", model="tsavage68/Summary_L3_1000steps_1e7rate_01beta_CSFTDPO")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("tsavage68/Summary_L3_1000steps_1e7rate_01beta_CSFTDPO")
model = AutoModelForCausalLM.from_pretrained("tsavage68/Summary_L3_1000steps_1e7rate_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/Summary_L3_1000steps_1e7rate_01beta_CSFTDPO with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tsavage68/Summary_L3_1000steps_1e7rate_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/Summary_L3_1000steps_1e7rate_01beta_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tsavage68/Summary_L3_1000steps_1e7rate_01beta_CSFTDPO

SGLang

How to use tsavage68/Summary_L3_1000steps_1e7rate_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/Summary_L3_1000steps_1e7rate_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/Summary_L3_1000steps_1e7rate_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/Summary_L3_1000steps_1e7rate_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/Summary_L3_1000steps_1e7rate_01beta_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use tsavage68/Summary_L3_1000steps_1e7rate_01beta_CSFTDPO with Docker Model Runner:
```
docker model run hf.co/tsavage68/Summary_L3_1000steps_1e7rate_01beta_CSFTDPO
```

Summary_L3_1000steps_1e7rate_01beta_CSFTDPO

This model is a fine-tuned version of tsavage68/Summary_L3_1000steps_1e7rate_SFT2 on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 0.5985
Rewards/chosen: 0.0302
Rewards/rejected: -0.6194
Rewards/accuracies: 0.1400
Rewards/margins: 0.6496
Logps/rejected: -21.4582
Logps/chosen: -9.0811
Logits/rejected: -1.1314
Logits/chosen: -1.1318

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-07
train_batch_size: 1
eval_batch_size: 1
seed: 42
gradient_accumulation_steps: 4
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
0.6896	0.2004	50	0.6887	0.0011	-0.0081	0.1300	0.0092	-15.3448	-9.3720	-1.0951	-1.0966
0.6884	0.4008	100	0.6748	0.0029	-0.0369	0.1400	0.0397	-15.6323	-9.3540	-1.0944	-1.0960
0.6591	0.6012	150	0.6445	0.0105	-0.1159	0.1400	0.1264	-16.4229	-9.2778	-1.0930	-1.0946
0.6351	0.8016	200	0.6267	0.0165	-0.1887	0.1400	0.2052	-17.1511	-9.2181	-1.0945	-1.0961
0.6358	1.0020	250	0.6157	0.0185	-0.2627	0.1400	0.2813	-17.8912	-9.1973	-1.0982	-1.0997
0.6306	1.2024	300	0.6088	0.0236	-0.3302	0.1400	0.3538	-18.5660	-9.1466	-1.1029	-1.1042
0.6303	1.4028	350	0.6051	0.0258	-0.3891	0.1400	0.4149	-19.1550	-9.1247	-1.1093	-1.1105
0.5829	1.6032	400	0.6023	0.0251	-0.4564	0.1400	0.4815	-19.8280	-9.1320	-1.1142	-1.1152
0.5941	1.8036	450	0.6007	0.0285	-0.5077	0.1400	0.5362	-20.3411	-9.0976	-1.1187	-1.1195
0.5754	2.0040	500	0.5999	0.0294	-0.5348	0.1400	0.5642	-20.6119	-9.0885	-1.1219	-1.1226
0.5759	2.2044	550	0.5994	0.0296	-0.5646	0.1400	0.5942	-20.9093	-9.0868	-1.1246	-1.1252
0.5575	2.4048	600	0.5990	0.0286	-0.5897	0.1400	0.6183	-21.1612	-9.0967	-1.1275	-1.1281
0.5235	2.6052	650	0.5987	0.0319	-0.6070	0.1400	0.6389	-21.3342	-9.0637	-1.1296	-1.1301
0.6277	2.8056	700	0.5986	0.0302	-0.6143	0.1400	0.6446	-21.4070	-9.0805	-1.1303	-1.1308
0.6079	3.0060	750	0.5985	0.0312	-0.6184	0.1400	0.6497	-21.4481	-9.0704	-1.1313	-1.1317
0.6422	3.2064	800	0.5985	0.0303	-0.6187	0.1400	0.6490	-21.4508	-9.0798	-1.1311	-1.1315
0.6589	3.4068	850	0.5985	0.0302	-0.6188	0.1400	0.6490	-21.4517	-9.0809	-1.1310	-1.1314
0.6247	3.6072	900	0.5986	0.0292	-0.6183	0.1400	0.6475	-21.4472	-9.0909	-1.1312	-1.1316
0.5393	3.8076	950	0.5985	0.0302	-0.6194	0.1400	0.6496	-21.4582	-9.0811	-1.1314	-1.1318
0.6252	4.0080	1000	0.5985	0.0302	-0.6194	0.1400	0.6496	-21.4582	-9.0811	-1.1314	-1.1318

Framework versions

Transformers 4.41.2
Pytorch 2.0.0+cu117
Datasets 2.20.0
Tokenizers 0.19.1

Downloads last month: 10

Safetensors

Model size

8B params

Tensor type

F16

Model tree for tsavage68/Summary_L3_1000steps_1e7rate_01beta_CSFTDPO

Base model

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

Finetuned

tsavage68/Summary_L3_1000steps_1e7rate_SFT2

Finetuned

(17)

this model