61 lines
1.6 KiB
Markdown
61 lines
1.6 KiB
Markdown
|
---
|
||
|
|
license: apache-2.0
|
||
|
|
base_model: google/functiongemma-270m-it
|
||
|
|
tags:
|
||
|
|
- function-calling
|
||
|
|
- asr
|
||
|
|
- bash
|
||
|
|
- voice-commands
|
||
|
|
- gemma
|
||
|
|
datasets:
|
||
|
|
- custom
|
||
|
|
language:
|
||
|
|
- en
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
---
|
||
|
|
|
||
|
|
# ASR-to-Bash (GGUF)
|
||
|
|
|
||
|
|
Fine-tuned FunctionGemma (270M) model that converts ASR (speech-to-text) transcriptions into executable bash commands.
|
||
|
|
|
||
|
|
## Usage
|
||
|
|
|
||
|
|
```python
|
||
|
|
# For llama.cpp / Ollama usage
|
||
|
|
# llama-cli -m asr-to-bash-q4_k_m.gguf -p 'Convert: list all files'
|
||
|
|
|
||
|
|
# Or with Python:
|
||
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||
|
|
|
||
|
|
model = AutoModelForCausalLM.from_pretrained("marksverdhai/asr-to-bash")
|
||
|
|
tokenizer = AutoTokenizer.from_pretrained("marksverdhai/asr-to-bash")
|
||
|
|
|
||
|
|
messages = [
|
||
|
|
{"role": "system", "content": "You are a helpful assistant that converts spoken commands into bash commands."},
|
||
|
|
{"role": "user", "content": "Convert this spoken command to bash: list all files including hidden ones"}
|
||
|
|
]
|
||
|
|
|
||
|
|
inputs = tokenizer.apply_chat_template(messages, return_tensors="pt")
|
||
|
|
outputs = model.generate(inputs, max_new_tokens=50)
|
||
|
|
print(tokenizer.decode(outputs[0]))
|
||
|
|
# Output: ls -la
|
||
|
|
```
|
||
|
|
|
||
|
|
## Examples
|
||
|
|
|
||
|
|
| ASR Transcription | Bash Command |
|
||
|
|
|------------------|--------------|
|
||
|
|
| "list all files" | `ls -la` |
|
||
|
|
| "git status" | `git status` |
|
||
|
|
| "change directory to home" | `cd ~` |
|
||
|
|
| "kill process one two three four" | `kill 1234` |
|
||
|
|
| "show running containers" | `docker ps` |
|
||
|
|
|
||
|
|
## Training
|
||
|
|
|
||
|
|
Fine-tuned using Unsloth with LoRA on a custom dataset of ~100 ASR transcription to bash command pairs.
|
||
|
|
|
||
|
|
- Base model: `google/functiongemma-270m-it`
|
||
|
|
- LoRA rank: 16
|
||
|
|
- Training epochs: 3
|