Upload Kaggle TPU pretrained tiny causal LM
Browse files- config.json +31 -0
- pytorch_model.bin +3 -0
- tokenizer.json +0 -0
- tokenizer_config.json +13 -0
- training_config.json +40 -0
config.json
ADDED
|
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation_function": "gelu_new",
|
| 3 |
+
"add_cross_attention": false,
|
| 4 |
+
"attn_pdrop": 0.0,
|
| 5 |
+
"bos_token_id": 50256,
|
| 6 |
+
"embd_pdrop": 0.0,
|
| 7 |
+
"eos_token_id": 50256,
|
| 8 |
+
"initializer_range": 0.02,
|
| 9 |
+
"layer_norm_epsilon": 1e-05,
|
| 10 |
+
"model_type": "gpt2",
|
| 11 |
+
"n_ctx": 2048,
|
| 12 |
+
"n_embd": 768,
|
| 13 |
+
"n_head": 12,
|
| 14 |
+
"n_inner": null,
|
| 15 |
+
"n_layer": 12,
|
| 16 |
+
"n_positions": 2048,
|
| 17 |
+
"pad_token_id": null,
|
| 18 |
+
"reorder_and_upcast_attn": false,
|
| 19 |
+
"resid_pdrop": 0.0,
|
| 20 |
+
"scale_attn_by_inverse_layer_idx": false,
|
| 21 |
+
"scale_attn_weights": true,
|
| 22 |
+
"summary_activation": null,
|
| 23 |
+
"summary_first_dropout": 0.1,
|
| 24 |
+
"summary_proj_to_labels": true,
|
| 25 |
+
"summary_type": "cls_index",
|
| 26 |
+
"summary_use_proj": true,
|
| 27 |
+
"tie_word_embeddings": true,
|
| 28 |
+
"transformers_version": "5.8.0",
|
| 29 |
+
"use_cache": false,
|
| 30 |
+
"vocab_size": 50257
|
| 31 |
+
}
|
pytorch_model.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:712620f9757e650a18a90a19d0a3fe7895cec55e742c8f33b4de14961245321c
|
| 3 |
+
size 655341431
|
tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
tokenizer_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|endoftext|>",
|
| 5 |
+
"eos_token": "<|endoftext|>",
|
| 6 |
+
"errors": "replace",
|
| 7 |
+
"is_local": false,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 1024,
|
| 10 |
+
"pad_token": "<|endoftext|>",
|
| 11 |
+
"tokenizer_class": "GPT2Tokenizer",
|
| 12 |
+
"unk_token": "<|endoftext|>"
|
| 13 |
+
}
|
training_config.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"dataset_name": "LLM360/TxT360",
|
| 3 |
+
"dataset_config": null,
|
| 4 |
+
"split": "train",
|
| 5 |
+
"text_column": "text",
|
| 6 |
+
"streaming": true,
|
| 7 |
+
"download_rows": null,
|
| 8 |
+
"shuffle_buffer": 50000,
|
| 9 |
+
"preprocessing_batch_size": 128,
|
| 10 |
+
"iterable_shards_when_downloaded": 1024,
|
| 11 |
+
"tokenizer_name": "gpt2",
|
| 12 |
+
"block_size": 2048,
|
| 13 |
+
"model_preset": "tiny_125m",
|
| 14 |
+
"n_layer": null,
|
| 15 |
+
"n_embd": null,
|
| 16 |
+
"n_head": null,
|
| 17 |
+
"resid_pdrop": 0.0,
|
| 18 |
+
"embd_pdrop": 0.0,
|
| 19 |
+
"attn_pdrop": 0.0,
|
| 20 |
+
"gradient_checkpointing": false,
|
| 21 |
+
"max_parameters": 600000000,
|
| 22 |
+
"num_tpu_processes": 1,
|
| 23 |
+
"per_device_batch_size": 8,
|
| 24 |
+
"gradient_accumulation_steps": 4,
|
| 25 |
+
"max_steps": 10000,
|
| 26 |
+
"learning_rate": 0.0003,
|
| 27 |
+
"weight_decay": 0.1,
|
| 28 |
+
"beta1": 0.9,
|
| 29 |
+
"beta2": 0.95,
|
| 30 |
+
"warmup_steps": 100,
|
| 31 |
+
"max_grad_norm": 1.0,
|
| 32 |
+
"num_workers": 0,
|
| 33 |
+
"seed": 42,
|
| 34 |
+
"log_every": 20,
|
| 35 |
+
"save_every": 100,
|
| 36 |
+
"output_dir": "/kaggle/working/tiny-lm-tpu",
|
| 37 |
+
"resume_from": null,
|
| 38 |
+
"push_to_hub": true,
|
| 39 |
+
"hub_model_id": "moos124/tiny-lm-125m"
|
| 40 |
+
}
|