Instructions to use spandyie/amadablam-322m-instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use spandyie/amadablam-322m-instruct with Transformers:

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

pipe = pipeline("text-generation", model="spandyie/amadablam-322m-instruct", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("spandyie/amadablam-322m-instruct", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use spandyie/amadablam-322m-instruct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "spandyie/amadablam-322m-instruct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "spandyie/amadablam-322m-instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/spandyie/amadablam-322m-instruct

SGLang

How to use spandyie/amadablam-322m-instruct 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 "spandyie/amadablam-322m-instruct" \
    --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": "spandyie/amadablam-322m-instruct",
		"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 "spandyie/amadablam-322m-instruct" \
        --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": "spandyie/amadablam-322m-instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use spandyie/amadablam-322m-instruct with Docker Model Runner:
```
docker model run hf.co/spandyie/amadablam-322m-instruct
```

Ama Dablam 322M Instruct

Instruction-tuned Ama Dablam: replies in whichever script you use — Devanagari, IAST, or phonetic romanized Nepali (e.g. "nepaalko raajdhaani kahaa ho?"). Fine-tuned on 259k Nepali instructions rendered 85/5/10 across the three scripts, with 5% pretraining replay to preserve base abilities.

response bpb (lower=better)	overall	Devanagari	IAST	phonetic
base (chat format)	0.570	0.524	1.008	1.171
instruct (this model)	0.347	0.315	0.696	0.741

Forgetting vs base after SFT: ne +0.011, mai +0.030, bho +0.039 bpb (all small); IAST-mai/bho drift +0.06–0.08 (uncovered combinations; documented).

Usage (chat template built in)

from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("spandyie/amadablam-322m-instruct")
model = AutoModelForCausalLM.from_pretrained("spandyie/amadablam-322m-instruct",
                                             trust_remote_code=True)
msgs = [{"role": "user", "content": "स्वस्थ रहन के गर्नुपर्छ?"}]
ids = tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt")
out = model.generate(ids, max_new_tokens=200)   # sampling defaults ship in generation_config
print(tok.decode(out[0][ids.shape[1]:], skip_special_tokens=True))

Limitations (read before use)

License: CC-BY-NC-4.0 (non-commercial) — the instruction data derives partly from Alpaca/Dolly-family sets carrying OpenAI-generated-content terms.
322M params: follows instructions and matches your script, but facts are unreliable — built for formatting/conversation, not QA accuracy.
Known failure mode: creative Devanagari prompts (poems, greetings) can degenerate into English text.
Multi-turn is lightly trained (template supports it; quality untested).
This model is cacheless by design. generation_config.json sets use_cache: false by default. If you override generation kwargs, do not set use_cache=True — it will silently produce incoherent, context-free output rather than an error (only the last token gets fed back each step).
Right-pad with BOS id 16368.
No system role; single-seed SFT; see the technical report for full evaluation.

Downloads last month: 25

Safetensors

Model size

0.3B params

Tensor type

F32

Model tree for spandyie/amadablam-322m-instruct

Base model

spandyie/amadablam-322m

Finetuned

(1)

this model