Instructions to use Polygl0t/GigaVerbo-v2-ablation-Synth-1.5B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Polygl0t/GigaVerbo-v2-ablation-Synth-1.5B with Transformers:

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

pipe = pipeline("text-generation", model="Polygl0t/GigaVerbo-v2-ablation-Synth-1.5B")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Polygl0t/GigaVerbo-v2-ablation-Synth-1.5B")
model = AutoModelForCausalLM.from_pretrained("Polygl0t/GigaVerbo-v2-ablation-Synth-1.5B", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Polygl0t/GigaVerbo-v2-ablation-Synth-1.5B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Polygl0t/GigaVerbo-v2-ablation-Synth-1.5B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Polygl0t/GigaVerbo-v2-ablation-Synth-1.5B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/Polygl0t/GigaVerbo-v2-ablation-Synth-1.5B

SGLang

How to use Polygl0t/GigaVerbo-v2-ablation-Synth-1.5B 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 "Polygl0t/GigaVerbo-v2-ablation-Synth-1.5B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Polygl0t/GigaVerbo-v2-ablation-Synth-1.5B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "Polygl0t/GigaVerbo-v2-ablation-Synth-1.5B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Polygl0t/GigaVerbo-v2-ablation-Synth-1.5B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use Polygl0t/GigaVerbo-v2-ablation-Synth-1.5B with Docker Model Runner:
```
docker model run hf.co/Polygl0t/GigaVerbo-v2-ablation-Synth-1.5B
```

GigaVerbo-v2-ablation-Synth-1.5B / training_config.yaml

nicholasKluge

Update training_config.yaml

f2ea872 verified 8 months ago

Raw

History Blame Contribute Delete

2.44 kB

	# Directory settings
	checkpoint_dir: "/lustre/scratch/data/polyglot_datasets/portuguese/checkpoints/models/gigaverbo-synth"
	train_dataset_dir:
	# Total: ~50B
	- /lustre/scratch/data/polyglot_datasets/portuguese/tokenized_synth
	- /lustre/scratch/data/polyglot_datasets/portuguese/tokenized_synth
	- /lustre/scratch/data/polyglot_datasets/portuguese/tokenized_synth
	- /lustre/scratch/data/polyglot_datasets/portuguese/tokenized_synth
	- /lustre/scratch/data/polyglot_datasets/portuguese/tokenized_synth
	val_dataset_dir: "/lustre/scratch/data/polyglot_datasets/portuguese/synth_val"
	dataset_type: "parquet"
	cache_dir: "/lustre/mlnvme/data/nanotronics/.cache"

	# Data loading settings
	pin_memory: true
	num_workers_for_dataloader: 16
	shuffle_dataset: true
	mask_eos_token: false
	mask_pad_token: false

	# Model architecture settings
	vocab_size: 49152
	num_hidden_layers: 28
	num_attention_heads: 16
	num_key_value_heads: 8
	head_dim: 128
	hidden_size: 2048
	intermediate_size: 6144
	max_position_embeddings: 4096
	tie_word_embeddings: true
	hidden_act: "silu"
	output_hidden_states: false
	attn_implementation: "flash_attention_2"
	use_cache: false
	no_rope_layer_interval: null
	rope_theta: 50000.0
	rope_scale_factor: null
	rms_norm_eps: 0.000001

	# Training settings
	total_batch_size: 2097152
	micro_batch_size: 8
	eval_micro_batch_size: 4
	num_train_epochs: 1
	warmup_steps: 2000
	max_learning_rate: 0.0008
	min_learning_rate: 0.0
	muon_learning_rate: 0.008
	weight_decay: 0.1
	beta1: 0.9
	beta2: 0.95
	eps: 0.00000001
	lr_decay_type: "cosine"
	use_sqrt: false
	lr_decay_iters_coef: 1.0
	seed: 1337
	max_steps: 22000
	max_grad_norm: 1.0

	# Precision and optimization settings
	torch_compile: false
	mat_mul_precision: "highest"
	tf32: true
	bf16: true
	gradient_checkpointing: false
	use_liger_kernel: true
	static_graph: false

	# Hub settings
	push_to_hub: false
	hub_token: null
	hub_model_id: null

	# Tokenizer and Reference model
	tokenizer_name_or_path: "/lustre/mlnvme/data/polyglot/portuguese/checkpoints/tokenizers/sp-bpe"
	chat_template_path: null
	reference_model: "HuggingFaceTB/SmolLM2-360M"
	continual_pretraining: false

	# Checkpoint settings
	resume_from_checkpoint: null
	checkpointing_steps: 2000
	begin_new_stage: false
	stage_name: "single_cosine"

	# Miscellaneous settings
	sanity_check: false
	sanity_check_num_samples: 100000
	wandb_token: null
	wandb_id: "gigaverbo-synth-ablation"
	wandb_project: "Polyglot"
	wandb_desc: "Developing LLMs for low-resource languages"