Instructions to use c-bone/CrystaLLM-pi_base with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use c-bone/CrystaLLM-pi_base with Transformers:

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

pipe = pipeline("text-generation", model="c-bone/CrystaLLM-pi_base")

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("c-bone/CrystaLLM-pi_base")
model = AutoModelForCausalLM.from_pretrained("c-bone/CrystaLLM-pi_base", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use c-bone/CrystaLLM-pi_base with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/c-bone/CrystaLLM-pi_base

SGLang

How to use c-bone/CrystaLLM-pi_base 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 "c-bone/CrystaLLM-pi_base" \
    --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": "c-bone/CrystaLLM-pi_base",
		"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 "c-bone/CrystaLLM-pi_base" \
        --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": "c-bone/CrystaLLM-pi_base",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use c-bone/CrystaLLM-pi_base with Docker Model Runner:
```
docker model run hf.co/c-bone/CrystaLLM-pi_base
```

c-bone commited on Sep 8

Commit

e37242a

verified ·

1 Parent(s): ae03244

Add the training config that produced this checkpoint

Browse files

Files changed (1) hide show

training_config.jsonc +72 -0

training_config.jsonc ADDED Viewed

	@@ -0,0 +1,72 @@

+{
+  // Data Arguments
+  //################
+  "dataset_HF": "c-bone/lematerial_clean",
+  "pretrained_tokenizer_dir": "HF-cif-tokenizer",
+  "context_length": 1024,
+  "dataset_streaming": false,
+  // Filters
+  "remove_CIFs_above_context": false,
+  "remove_CIFs_with_unk": true,
+  // Conditional Arguments
+  //#######################
+  "activate_conditionality": "None",
+  // Model Arguments
+  //#################
+  // Model Depth
+  "n_embd": 512,
+  "n_layer": 8,
+  "n_head": 8,
+  // Dropout
+  "residual_dropout": 0.1,
+  "embedding_dropout": 0.1,
+  "attention_dropout": 0.1,
+  // Trainer Arguments
+  //###################
+  // Batching
+  "train_batch_size": 32,
+  "eval_batch_size": 32,
+  "gradient_accumulation_steps": 1,
+  "auto_find_batch_size": false,
+  // Learning Rate and Optimizer
+  "learning_rate": 0.001,
+  "lr_scheduler_type": "cosine_with_min_lr",
+  "lr_scheduler_kwargs": {
+    "min_lr_rate": 0.001
+  },
+  "warmup_steps": 0,
+  "adam_beta1": 0.85,
+  "adam_beta2": 0.98,
+  "grad_clip": 1.0,
+  "weight_decay": 0.1,
+  // Logging
+  "output_dir": "model_ckpts/mpdb-small-base-lematerial",
+  "save_total_limit": 2,
+  "report_to": "wandb",
+  "eval_strategy": "steps",
+  "eval_steps": 5000,
+  "logging_steps": 500,
+  "save_strategy": "steps",
+  "max_steps": 1250000,
+  "early_stopping_patience": 10,
+  "early_stopping_threshold": 0.00001,
+  // Utils
+  // 42 is what the released checkpoint's training_args.bin records: this run predates the
+  // seed being passed through from the config, so the Trainer default applied.
+  "seed": 42,
+  "data_seed": 1,
+  "load_best_model_at_end": true,
+  "metric_for_best_model": "eval_loss",
+  "greater_is_better": false,
+  "torch_compile": true,
+  "fp16": true,
+  "deepspeed_config": "_config_files/deepspeed_default.json",
+  // CodeCarbon Arguments
+  //######################
+  "codecarbon": true,
+  "tracker_project": "CrystaLLM-pi"
+}