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)

# 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=40)
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 / pretrain_llama_3b.sh

aapot

Add easylm training code

5a63fc6 over 2 years ago

Raw

History Blame

2.3 kB

	#! /bin/bash

	# Put your WANDB API key here to enable logging to wandb.
	export WANDB_API_KEY=''

	# TPU specific flags to improve training throughput
	export LIBTPU_INIT_ARGS='--xla_jf_spmd_threshold_for_windowed_einsum_mib=0 --xla_tpu_spmd_threshold_for_allgather_cse=10000 --xla_tpu_spmd_rewrite_einsum_with_reshape=true --xla_enable_async_all_gather=true --jax_enable_async_collective_offload=true --xla_tpu_enable_latency_hiding_scheduler=true TPU_MEGACORE=MEGACORE_DENSE'


	python3 -m EasyLM.models.llama.llama_train \
	--jax_distributed.initialize_jax_distributed=True \
	--mesh_dim='1,-1,1' \
	--dtype='bf16' \
	--total_steps=900000 \
	--eval_freq=50000 \
	--log_freq=1000 \
	--save_model_freq=2000 \
	--save_milestone_freq=50000 \
	--load_llama_config='3b' \
	--update_llama_config='' \
	--load_dataset_state='' \
	--load_checkpoint='' \
	--tokenizer.pretrained_model_name_or_path='./' \
	--optimizer.type='lion' \
	--optimizer.lion_optimizer.weight_decay=1.0 \
	--optimizer.lion_optimizer.lr_schedule_type='warmup_constant' \
	--optimizer.lion_optimizer.lr=3e-4 \
	--optimizer.lion_optimizer.end_lr=3e-5 \
	--optimizer.lion_optimizer.lr_warmup_steps=60000 \
	--optimizer.lion_optimizer.lr_decay_steps=100000 \
	--optimizer.lion_optimizer.bf16_momentum=True \
	--train_dataset.type='huggingface' \
	--train_dataset.text_processor.fields='text' \
	--train_dataset.huggingface_dataset.path='/researchdisk/lm_training_dataset_first_stage' \
	--train_dataset.huggingface_dataset.split='train' \
	--train_dataset.huggingface_dataset.seq_length=2048 \
	--train_dataset.huggingface_dataset.batch_size=64 \
	--eval_dataset.type='huggingface' \
	--eval_dataset.text_processor.fields='text' \
	--eval_dataset.huggingface_dataset.path='/researchdisk/lm_training_dataset_first_stage' \
	--eval_dataset.huggingface_dataset.split='validation' \
	--eval_dataset.huggingface_dataset.seq_length=2048 \
	--eval_dataset.huggingface_dataset.batch_size=64 \
	--checkpointer.save_optimizer_state=True \
	--logger.online=True \
	--logger.prefix='EasyLM' \
	--logger.project="llama-3b-finnish-v2" \
	--logger.output_dir="gs://finnish-nlp-research-us/llama-3b-v2-checkpoint" \
	--logger.wandb_dir="./"