Instructions to use Finnish-NLP/Ahma-3B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Finnish-NLP/Ahma-3B with Transformers:

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

pipe = pipeline("text-generation", model="Finnish-NLP/Ahma-3B")
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("Finnish-NLP/Ahma-3B")
model = AutoModelForCausalLM.from_pretrained("Finnish-NLP/Ahma-3B", 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 Finnish-NLP/Ahma-3B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Finnish-NLP/Ahma-3B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Finnish-NLP/Ahma-3B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Finnish-NLP/Ahma-3B

SGLang

How to use Finnish-NLP/Ahma-3B 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 "Finnish-NLP/Ahma-3B" \
    --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": "Finnish-NLP/Ahma-3B",
		"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 "Finnish-NLP/Ahma-3B" \
        --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": "Finnish-NLP/Ahma-3B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Finnish-NLP/Ahma-3B with Docker Model Runner:
```
docker model run hf.co/Finnish-NLP/Ahma-3B
```

Ahma-3B / EasyLM /scripts /diff_checkpoint.py

aapot

Add easylm training code

5a63fc6 over 2 years ago

Raw History Blame Contribute Delete

1.85 kB

# This script converts model checkpoint trained by EsayLM to a standard
# mspack checkpoint that can be loaded by huggingface transformers or
# flax.serialization.msgpack_restore. Such conversion allows models to be
# used by other frameworks that integrate with huggingface transformers.

import pprint
from functools import partial
import os
import numpy as np
import jax
import jax.numpy as jnp
import flax.serialization
import mlxu
from EasyLM.checkpoint import StreamingCheckpointer
from EasyLM.jax_utils import float_to_dtype


FLAGS, FLAGS_DEF = mlxu.define_flags_with_default(
    recover_diff=False,
    load_base_checkpoint='',
    load_target_checkpoint='',
    output_file='',
    streaming=True,
    float_dtype='bf16',
)


def main(argv):
    assert FLAGS.load_base_checkpoint != '' and FLAGS.load_target_checkpoint != ''
    assert FLAGS.output_file != ''
    base_params = StreamingCheckpointer.load_trainstate_checkpoint(
        FLAGS.load_base_checkpoint, disallow_trainstate=True
    )[1]['params']

    target_params = StreamingCheckpointer.load_trainstate_checkpoint(
        FLAGS.load_target_checkpoint, disallow_trainstate=True
    )[1]['params']

    if FLAGS.recover_diff:
        params = jax.tree_util.tree_map(
            lambda b, t: b + t, base_params, target_params
        )
    else:
        params = jax.tree_util.tree_map(
            lambda b, t: t - b, base_params, target_params
        )

    if FLAGS.streaming:
        StreamingCheckpointer.save_train_state_to_file(
            params, FLAGS.output_file, float_dtype=FLAGS.float_dtype
        )
    else:
        params = float_to_dtype(params, FLAGS.float_dtype)
        with mlxu.open_file(FLAGS.output, 'wb') as fout:
            fout.write(flax.serialization.msgpack_serialize(params, in_place=True))


if __name__ == "__main__":
    mlxu.run(main)

	# This script converts model checkpoint trained by EsayLM to a standard
	# mspack checkpoint that can be loaded by huggingface transformers or
	# flax.serialization.msgpack_restore. Such conversion allows models to be
	# used by other frameworks that integrate with huggingface transformers.

	import pprint
	from functools import partial
	import os
	import numpy as np
	import jax
	import jax.numpy as jnp
	import flax.serialization
	import mlxu
	from EasyLM.checkpoint import StreamingCheckpointer
	from EasyLM.jax_utils import float_to_dtype


	FLAGS, FLAGS_DEF = mlxu.define_flags_with_default(
	recover_diff=False,
	load_base_checkpoint='',
	load_target_checkpoint='',
	output_file='',
	streaming=True,
	float_dtype='bf16',
	)


	def main(argv):
	assert FLAGS.load_base_checkpoint != '' and FLAGS.load_target_checkpoint != ''
	assert FLAGS.output_file != ''
	base_params = StreamingCheckpointer.load_trainstate_checkpoint(
	FLAGS.load_base_checkpoint, disallow_trainstate=True
	)[1]['params']

	target_params = StreamingCheckpointer.load_trainstate_checkpoint(
	FLAGS.load_target_checkpoint, disallow_trainstate=True
	)[1]['params']

	if FLAGS.recover_diff:
	params = jax.tree_util.tree_map(
	lambda b, t: b + t, base_params, target_params
	)
	else:
	params = jax.tree_util.tree_map(
	lambda b, t: t - b, base_params, target_params
	)

	if FLAGS.streaming:
	StreamingCheckpointer.save_train_state_to_file(
	params, FLAGS.output_file, float_dtype=FLAGS.float_dtype
	)
	else:
	params = float_to_dtype(params, FLAGS.float_dtype)
	with mlxu.open_file(FLAGS.output, 'wb') as fout:
	fout.write(flax.serialization.msgpack_serialize(params, in_place=True))


	if __name__ == "__main__":
	mlxu.run(main)