zchen/b2txt25
1
0
Fork 0
Code Issues Pull Requests 1 Actions Packages Projects Releases Wiki Activity

additional documentation

Browse Source
This commit is contained in:
nckcard
2025-07-02 14:28:34 -07:00
parent b2d25bd606
commit d87462df51
6 changed files with 48 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
language_model/README.md
View File

@@ -23,18 +23,26 @@ The `language-model-standalone.py` script included here is made to work with the
3. Connect to the `localhost` redis server (or a different server, specified by the `--redis_ip` and `--redis_port` args)
4. Wait to receive phoneme logits via redis, and then make word predictions and pass them back via redis.
### `language-model-standalone.py` input args
See the bottom of the `language-model-standalone.py` script for a full list of input args.
### run a 1gram model
To run the 1gram language model from the root directory of this repository:
```bash
conda activate b2txt_lm
python language_model/language-model-standalone.py --lm_path language_model/pretrained_language_models/openwebtext_1gram_lm_sil --do_opt --nbest 100 --acoustic_scale 0.325 --blank_penalty 90 --alpha 0.55 --redis_ip localhost --gpu_number 0
```
### run a 3gram model
To run the 3gram language model from the root directory of this repository (requires ~60GB RAM):
```bash
conda activate b2txt_lm
python language_model/language-model-standalone.py --lm_path language_model/pretrained_language_models/openwebtext_3gram_lm_sil --do_opt --nbest 100 --acoustic_scale 0.325 --blank_penalty 90 --alpha 0.55 --redis_ip localhost --gpu_number 0
```
### run a 5gram model
To run the 5gram language model from the root directory of this repository (requires ~300GB of RAM):
```bash
conda activate b2txt_lm