mirror of
https://github.com/pese-git/llm-arch-research.git
synced 2026-01-23 21:10:54 +00:00
cfb4b6dfb1
- Add core Gemma model (architecture, attention, GeGLU, RoPE, RMSNorm, etc) - Add configs for training and generation: gemma_train.json, gemma_generate.json - Add Gemma notebook for exploratory analysis and demonstration - Add __init__.py for Gemma submodule - Update run_llm_experiment.py to support Gemma experiment configs test(gemma): add comprehensive unit tests for Gemma - Test forward pass (with/without cache) - Test autoregressive generation (greedy, top-k, top-p) - Test shape correctness and max sequence length errors - Test multi-layer stack and token embeddings docs: add documentation notebook for Gemma usage and analysis Closes: #issue (if applicable)
28 lines
822 B
JSON
28 lines
822 B
JSON
{
|
|
"bpe_tokenizer": "checkpoints/bpe_tokenizer.json",
|
|
"bpe_vocab_size": 1000,
|
|
"bpe_special_tokens": ["<pad>", "<unk>", "<bos>", "<eos>"],
|
|
"test_prompts": ["Open source AI", "What is Llama?"],
|
|
"model_config": {
|
|
"vocab_size": null,
|
|
"embed_dim": 256,
|
|
"num_q_heads": 4,
|
|
"num_kv_heads": 2,
|
|
"head_size": 64,
|
|
"num_layers": 4,
|
|
"max_position_embeddings": 512,
|
|
"num_experts": 8,
|
|
"top_k_experts": 2,
|
|
"window_size": 16,
|
|
"dropout": 0.1
|
|
},
|
|
"model_weights": "checkpoints/gemma-bpe/model.pt",
|
|
"model_config_path": "checkpoints/gemma-bpe/config.json",
|
|
"training": {
|
|
"learning_rate": 0.0003,
|
|
"batch_size": 2,
|
|
"num_epochs": 3,
|
|
"warmup_steps": 50
|
|
},
|
|
"log_path": "checkpoints/gemma_only_training_logs.json"
|
|
} |