Instructions to use kuluruvineeth/manas-64m-full-dist with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use kuluruvineeth/manas-64m-full-dist with Transformers:

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

pipe = pipeline("text-generation", model="kuluruvineeth/manas-64m-full-dist")
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("kuluruvineeth/manas-64m-full-dist")
model = AutoModelForCausalLM.from_pretrained("kuluruvineeth/manas-64m-full-dist", 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 kuluruvineeth/manas-64m-full-dist with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "kuluruvineeth/manas-64m-full-dist"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "kuluruvineeth/manas-64m-full-dist",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/kuluruvineeth/manas-64m-full-dist

SGLang

How to use kuluruvineeth/manas-64m-full-dist 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 "kuluruvineeth/manas-64m-full-dist" \
    --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": "kuluruvineeth/manas-64m-full-dist",
		"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 "kuluruvineeth/manas-64m-full-dist" \
        --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": "kuluruvineeth/manas-64m-full-dist",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use kuluruvineeth/manas-64m-full-dist with Docker Model Runner:
```
docker model run hf.co/kuluruvineeth/manas-64m-full-dist
```

Manas-64M · Distillation

The dense SFT model distilled from the mixture-of-experts SFT model: it learns from the teacher's full next-token distribution as well as from the data.

Try it: chat Space · Every stage: collection · Code: github.com/kuluruvineeth/manas · Data: manas_dataset

Results

	value
held-out loss	1.131
training loss, first → last log point	1.322 → 1.116
training steps	3,000
tokens processed	≈50M (throughput × time)
wall-clock	10 min at ~86k tokens/s

Benchmarks

Zero-shot multiple choice, 500 items per task, scored by comparing answer log-likelihoods (scripts/evaluate.py).

benchmark	accuracy	random chance	items
arc_challenge	19.2%	25.1%	500
arc_easy	33.4%	25.0%	500
hellaswag	34.4%	25.0%	500
mmlu	25.2%	25.0%	500
openbookqa	23.8%	25.0%	500
average	27.2%

A 64M model has far less capacity than these benchmarks assume, so scores sit near random chance. They are published because hiding them would be dishonest, and they move by about a point between stages, which is within noise at 500 items.

Use

from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "kuluruvineeth/manas-64m-full-dist"
tokenizer = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(repo)

messages = [{"role": "user", "content": "What is the capital of France?"}]
inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt", return_dict=True)
out = model.generate(**inputs, max_new_tokens=60, do_sample=False)
print(tokenizer.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

The export is a standard transformers checkpoint, so vLLM and SGLang load it with no custom code (vllm serve kuluruvineeth/manas-64m-full-dist).

Training

Lineage: starts from manas-64m-full-sft.
Method: 0.5 × cross-entropy + 0.5 × KL to the teacher at temperature 1.5, teacher manas-64m-full-sft-moe.
Data: sft_t2t_mini.jsonl (sources and licenses).
Trainer: trainer/train_distillation.py, peak learning rate 5e-6 (from the training log).
Hardware: one A100 or H100 GPU on Modal (gpu/modal_train.py).
Architecture: Qwen3 layout: 768 hidden, 8 layers, 8 query / 4 key-value heads (head dim 96), SwiGLU MLP of width 2,432, RoPE θ = 10⁶, tied embeddings, 6,400-token byte-level BPE vocabulary. 63,912,192 parameters.

Limitations

English only. At this size the model has little world knowledge: it is confidently wrong about facts that need it, loses the thread in long conversations, and should not be used for anything that matters. It is a teaching and research artifact for studying how each training stage changes a model.

Files

model.safetensors, config.json, tokenizer.json, chat_template.jinja: the transformers export
full_dist_768.pth: the raw training checkpoint for the training code
full_dist_768_metrics.jsonl, full_dist_768_curves.png: the training log and its plot