Upload README.md with huggingface_hub
Browse files
README.md
ADDED
@@ -0,0 +1,65 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
base_model: Maykeye/TinyLLama-v0
|
3 |
+
inference: false
|
4 |
+
license: apache-2.0
|
5 |
+
model_creator: Maykeye
|
6 |
+
model_name: TinyLLama-v0
|
7 |
+
pipeline_tag: text-generation
|
8 |
+
quantized_by: afrideva
|
9 |
+
tags:
|
10 |
+
- gguf
|
11 |
+
- ggml
|
12 |
+
- quantized
|
13 |
+
- q2_k
|
14 |
+
- q3_k_m
|
15 |
+
- q4_k_m
|
16 |
+
- q5_k_m
|
17 |
+
- q6_k
|
18 |
+
- q8_0
|
19 |
+
---
|
20 |
+
# Maykeye/TinyLLama-v0-GGUF
|
21 |
+
|
22 |
+
Quantized GGUF model files for [TinyLLama-v0](https://huggingface.co/Maykeye/TinyLLama-v0) from [Maykeye](https://huggingface.co/Maykeye)
|
23 |
+
|
24 |
+
|
25 |
+
| Name | Quant method | Size |
|
26 |
+
| ---- | ---- | ---- |
|
27 |
+
| [tinyllama-v0.fp16.gguf](https://huggingface.co/afrideva/TinyLLama-v0-GGUF/resolve/main/tinyllama-v0.fp16.gguf) | fp16 | 11.08 MB |
|
28 |
+
| [tinyllama-v0.q2_k.gguf](https://huggingface.co/afrideva/TinyLLama-v0-GGUF/resolve/main/tinyllama-v0.q2_k.gguf) | q2_k | 5.47 MB |
|
29 |
+
| [tinyllama-v0.q3_k_m.gguf](https://huggingface.co/afrideva/TinyLLama-v0-GGUF/resolve/main/tinyllama-v0.q3_k_m.gguf) | q3_k_m | 5.63 MB |
|
30 |
+
| [tinyllama-v0.q4_k_m.gguf](https://huggingface.co/afrideva/TinyLLama-v0-GGUF/resolve/main/tinyllama-v0.q4_k_m.gguf) | q4_k_m | 5.79 MB |
|
31 |
+
| [tinyllama-v0.q5_k_m.gguf](https://huggingface.co/afrideva/TinyLLama-v0-GGUF/resolve/main/tinyllama-v0.q5_k_m.gguf) | q5_k_m | 5.95 MB |
|
32 |
+
| [tinyllama-v0.q6_k.gguf](https://huggingface.co/afrideva/TinyLLama-v0-GGUF/resolve/main/tinyllama-v0.q6_k.gguf) | q6_k | 6.72 MB |
|
33 |
+
| [tinyllama-v0.q8_0.gguf](https://huggingface.co/afrideva/TinyLLama-v0-GGUF/resolve/main/tinyllama-v0.q8_0.gguf) | q8_0 | 6.75 MB |
|
34 |
+
|
35 |
+
|
36 |
+
|
37 |
+
## Original Model Card:
|
38 |
+
This is a first version of recreating roneneldan/TinyStories-1M but using Llama architecture.
|
39 |
+
|
40 |
+
* Full training process is included in the notebook train.ipynb. Recreating it as simple as downloading
|
41 |
+
TinyStoriesV2-GPT4-train.txt and TinyStoriesV2-GPT4-valid.txt in the same folder with the notebook and running
|
42 |
+
the cells. Validation content is not used by the script so you put anythin in
|
43 |
+
|
44 |
+
* Backup directory has a script do_backup that I used to copy weights from remote machine to local.
|
45 |
+
Weight are generated too quickly, so by the time script copied weihgt N+1
|
46 |
+
|
47 |
+
* This is extremely PoC version. Training truncates stories that are longer than context size and doesn't use
|
48 |
+
any sliding window to train story not from the start
|
49 |
+
|
50 |
+
* Training took approximately 9 hours (3 hours per epoch) on 40GB A100. ~30GB VRAM was used
|
51 |
+
|
52 |
+
* I use tokenizer from open_llama_3b. However I had troubles with it locally(https://github.com/openlm-research/open_llama/issues/69).
|
53 |
+
I had no troubles on the cloud machine with preninstalled libraries.
|
54 |
+
|
55 |
+
* Demo script is demo.py
|
56 |
+
|
57 |
+
* Validation script is provided: valid.py. use it like `python valid.py path/to/TinyStoriesV2-GPT4-valid.txt [optional-model-id-or-path]`:
|
58 |
+
After training I decided that it's not necessary to beat validation into chunks
|
59 |
+
|
60 |
+
* Also this version uses very stupid caching mechinsm to shuffle stories for training: it keeps cache of N recently loaded chunks
|
61 |
+
so if random shuffle asks for a story, it may use cache or load chunk.
|
62 |
+
Training dataset is too small, so in next versions I will get rid of it.
|
63 |
+
|
64 |
+
|
65 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|