schihei/llama2.c
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #5 from danielgross/pleasantify-dx

Browse Source 
Make sample.py work out of the box
This commit is contained in:
Andrej
2023-07-23 11:58:03 -07:00
committed by GitHub
parent 3b7b4878b4 8c383c28f9
commit 7d401d530c
2 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+2 -2
View File
@@ -80,7 +80,7 @@ But note that this only emits the SentencePiece tokens. To decode the tokens int
python run_wrap.py python run_wrap.py
``` ```
Watch the tokens stream by, fun! We can also run the PyTorch inference script for comparison: Watch the tokens stream by, fun! We can also run the PyTorch inference script for comparison (to run, add [model.ckpt](https://drive.google.com/file/d/1SM0rMxzy7babB-v4MfTg1GFqOCgWar5w/view?usp=share_link) to /out if you haven't already):
```bash ```bash
python sample.py python sample.py
@@ -92,7 +92,7 @@ Which gives the same results. More detailed testing will be done in `test_all.py
$ pytest $ pytest
``` ```
Currently you will need two files to run the test: the [model.bin](https://drive.google.com/file/d/1aTimLdx3JktDXxcHySNrZJOOk8Vb1qBR/view?usp=share_link) file and the [model.ckpt](https://drive.google.com/file/d/1SM0rMxzy7babB-v4MfTg1GFqOCgWar5w/view?usp=share_link) file from PyTorch training I ran earlier. I have to think through running the tests without having to download 200MB of data. Currently you will need two files to test or sample: the [model.bin](https://drive.google.com/file/d/1aTimLdx3JktDXxcHySNrZJOOk8Vb1qBR/view?usp=share_link) file and the [model.ckpt](https://drive.google.com/file/d/1SM0rMxzy7babB-v4MfTg1GFqOCgWar5w/view?usp=share_link) file from PyTorch training I ran earlier. I have to think through running the tests without having to download 200MB of data.
## unsorted todos ## unsorted todos
sample.py
+1 -1
View File
@@ -17,7 +17,7 @@ max_new_tokens = 100 # number of tokens generated in each sample
temperature = 1.0 # 1.0 = no change, < 1.0 = less random, > 1.0 = more random, in predictions temperature = 1.0 # 1.0 = no change, < 1.0 = less random, > 1.0 = more random, in predictions
top_k = 300 # retain only the top_k most likely tokens, clamp others to have 0 probability top_k = 300 # retain only the top_k most likely tokens, clamp others to have 0 probability
seed = 1337 seed = 1337
device = 'cuda' # examples: 'cpu', 'cuda', 'cuda:0', 'cuda:1', etc. device = 'cuda' if torch.cuda.is_available() else 'cpu' # examples: 'cpu', 'cuda', 'cuda:0', 'cuda:1', etc.
#dtype = 'bfloat16' if torch.cuda.is_available() and torch.cuda.is_bf16_supported() else 'float16' # 'float32' or 'bfloat16' or 'float16' #dtype = 'bfloat16' if torch.cuda.is_available() and torch.cuda.is_bf16_supported() else 'float16' # 'float32' or 'bfloat16' or 'float16'
dtype = "float32" dtype = "float32"
compile = False # use PyTorch 2.0 to compile the model to be faster compile = False # use PyTorch 2.0 to compile the model to be faster