Upload folder using huggingface_hub
Browse files- README.md +10 -0
- config.json +27 -0
- generation_config.json +9 -0
- model.safetensors +3 -0
- tokenizer.json +0 -0
- tokenizer_config.json +16 -0
README.md
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language: en
|
| 3 |
+
license: apache-2.0
|
| 4 |
+
tags:
|
| 5 |
+
- fwkv
|
| 6 |
+
- rosa
|
| 7 |
+
---
|
| 8 |
+
# FWKV-56M + ROSA (chat)
|
| 9 |
+
Factorised tied weights, vectorized decayed-accumulator recurrence, RWKV-8 ROSA copy signal, chat-tuned on HuggingFaceH4/ultrachat_200k.
|
| 10 |
+
**Best val perplexity:** 67.78
|
config.json
ADDED
|
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"accum_steps": 4,
|
| 3 |
+
"architectures": [
|
| 4 |
+
"FWKVLanguageModel"
|
| 5 |
+
],
|
| 6 |
+
"batch_size": 8,
|
| 7 |
+
"ce_chunk": 512,
|
| 8 |
+
"d_emb": 128,
|
| 9 |
+
"d_model": 512,
|
| 10 |
+
"dataset_name": "HuggingFaceH4/ultrachat_200k",
|
| 11 |
+
"dtype": "float32",
|
| 12 |
+
"epochs": 2,
|
| 13 |
+
"ffn_mult": 4,
|
| 14 |
+
"grad_clip": 1.0,
|
| 15 |
+
"lr": 0.0003,
|
| 16 |
+
"max_train_examples": 20000,
|
| 17 |
+
"max_val_examples": 1000,
|
| 18 |
+
"model_type": "fwkv",
|
| 19 |
+
"n_layers": 14,
|
| 20 |
+
"seq_len": 1024,
|
| 21 |
+
"tie_word_embeddings": true,
|
| 22 |
+
"train_split": "train_sft",
|
| 23 |
+
"transformers_version": "5.0.0",
|
| 24 |
+
"val_split": "test_sft",
|
| 25 |
+
"vocab_size": 50259,
|
| 26 |
+
"wkv_floor": 0.1
|
| 27 |
+
}
|
generation_config.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"do_sample": true,
|
| 3 |
+
"eos_token_id": 50256,
|
| 4 |
+
"max_new_tokens": 150,
|
| 5 |
+
"pad_token_id": 50256,
|
| 6 |
+
"temperature": 0.8,
|
| 7 |
+
"top_k": 50,
|
| 8 |
+
"transformers_version": "5.0.0"
|
| 9 |
+
}
|
model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6e8f6d2311298ada78470bf6054e0ba9046ebfb4b9bf0ad61b4b0de9eaeebaee
|
| 3 |
+
size 228050840
|
tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
tokenizer_config.json
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|endoftext|>",
|
| 5 |
+
"eos_token": "<|endoftext|>",
|
| 6 |
+
"errors": "replace",
|
| 7 |
+
"extra_special_tokens": [
|
| 8 |
+
"<|user|>",
|
| 9 |
+
"<|assistant|>"
|
| 10 |
+
],
|
| 11 |
+
"is_local": false,
|
| 12 |
+
"model_max_length": 1024,
|
| 13 |
+
"pad_token": "<|endoftext|>",
|
| 14 |
+
"tokenizer_class": "GPT2Tokenizer",
|
| 15 |
+
"unk_token": "<|endoftext|>"
|
| 16 |
+
}
|