Instructions to use Sudhanshu1985/slm-125m-raft with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Sudhanshu1985/slm-125m-raft with Transformers:

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

pipe = pipeline("text-generation", model="Sudhanshu1985/slm-125m-raft")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Sudhanshu1985/slm-125m-raft")
model = AutoModelForCausalLM.from_pretrained("Sudhanshu1985/slm-125m-raft", 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=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Sudhanshu1985/slm-125m-raft with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Sudhanshu1985/slm-125m-raft"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Sudhanshu1985/slm-125m-raft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Sudhanshu1985/slm-125m-raft

SGLang

How to use Sudhanshu1985/slm-125m-raft 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 "Sudhanshu1985/slm-125m-raft" \
    --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": "Sudhanshu1985/slm-125m-raft",
		"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 "Sudhanshu1985/slm-125m-raft" \
        --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": "Sudhanshu1985/slm-125m-raft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Sudhanshu1985/slm-125m-raft with Docker Model Runner:
```
docker model run hf.co/Sudhanshu1985/slm-125m-raft
```

slm-125m-raft

A RAFT (Retrieval-Augmented Fine-Tuning) version of the 125M legal/financial model: thesreedath/slm-125m-base fine-tuned to answer from a retrieved set of documents (one golden chunk + distractors) and to refuse when the answer is not present in any of them.

Companion to Sudhanshu1985/slm-125m-sft (plain grounded-QA SFT). RAFT adds distractor documents and a 30% "no-golden" negative rate, so the model learns to ignore irrelevant retrieved chunks and not hallucinate.

Results (held-out validation)

	Base	RAFT
Loss	2.81	0.85
Perplexity	16.54	2.34

(Val set is the harder RAFT set — multi-document with distractors.)

Training data

Sudhanshu1985/slm-125m-raft-dataset — 23,830 examples derived from the reviewed QA set:

70% positive: question + [golden chunk + 3 distractors] (shuffled) → grounded answer
30% negative: golden removed / unanswerable → "That is not stated in the context."
~200-token chunks so golden + distractors fit the 1024 window.

Recipe: 3 epochs, 1xH100, lr 2e-5 cosine, AdamW, bf16, effective batch 32, loss on the assistant span only.

Chat template

<|bos|><|system|>SYSTEM<|user|>USER<|assistant|>ANSWER<|eos|>

The user turn holds the retrieved documents followed by the question, e.g.:

Context:
[Document 1]
...
[Document 2]
...

<your question>

from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("Sudhanshu1985/slm-125m-raft")
model = AutoModelForCausalLM.from_pretrained("Sudhanshu1985/slm-125m-raft")

Limitations

Knows no facts of its own; works only over the documents you supply.
1024-token window — keep the retrieved doc set short.
Domain-biased toward US legal/financial register. Not legal or financial advice.

Downloads last month: 15

Safetensors

Model size

0.1B params

Tensor type

BF16

Model tree for Sudhanshu1985/slm-125m-raft

Base model

thesreedath/slm-125m-base

Finetuned

(14)

this model

Sudhanshu1985
/

slm-125m-raft