Sebastien De Greef
commited on
Commit
·
909b9b6
1
Parent(s):
6baccb3
handle push_to_hub_gguf and inference
Browse files
app.py
CHANGED
|
@@ -92,7 +92,20 @@ def load_data(dataset_name, data_template_style, data_template):
|
|
| 92 |
dataset = dataset.map(lambda examples: formatting_prompts_func(examples, data_template), batched=True)
|
| 93 |
return f"Data loaded {len(dataset)} records loaded.", gr.update(visible=True, interactive=True), gr.update(visible=True, interactive=True)
|
| 94 |
|
| 95 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 96 |
|
| 97 |
|
| 98 |
async def train_model(model_name: str, lora_r: int, lora_alpha: int, lora_dropout: float, per_device_train_batch_size: int, warmup_steps: int, max_steps: int,
|
|
@@ -143,9 +156,35 @@ async def train_model(model_name: str, lora_r: int, lora_alpha: int, lora_dropou
|
|
| 143 |
trainer.train()
|
| 144 |
return "Model training",gr.update(visible=True, interactive=False), gr.update(visible=True, interactive=True), gr.update(interactive=True)
|
| 145 |
|
| 146 |
-
def save_model():
|
| 147 |
-
|
| 148 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 149 |
|
| 150 |
# Create the Gradio interface
|
| 151 |
with gr.Blocks() as demo:
|
|
@@ -171,7 +210,7 @@ with gr.Blocks() as demo:
|
|
| 171 |
dataset_name = gr.Textbox(label="Dataset Name", value="yahma/alpaca-cleaned")
|
| 172 |
data_template_style = gr.Dropdown(label="Template", choices=["alpaca","custom"], value="alpaca", allow_custom_value=True)
|
| 173 |
with gr.Row():
|
| 174 |
-
|
| 175 |
|
| 176 |
### Instruction:
|
| 177 |
{}
|
|
@@ -184,7 +223,7 @@ with gr.Blocks() as demo:
|
|
| 184 |
gr.Markdown("---")
|
| 185 |
output_load_data = gr.Textbox(label="Data Load Status", value="Data not loaded", interactive=False)
|
| 186 |
load_data_btn = gr.Button("Load Dataset", interactive=True)
|
| 187 |
-
load_data_btn.click(load_data, inputs=[dataset_name, data_template_style,
|
| 188 |
|
| 189 |
with gr.Tab("Fine-Tuning"):
|
| 190 |
gr.Markdown("""### Fine-Tuned Model Parameters""")
|
|
@@ -238,18 +277,18 @@ with gr.Blocks() as demo:
|
|
| 238 |
with gr.Column():
|
| 239 |
merge_16bit = gr.Checkbox(label="Merge to 16bit", value=False, interactive=True)
|
| 240 |
merge_4bit = gr.Checkbox(label="Merge to 4bit", value=False, interactive=True)
|
| 241 |
-
|
| 242 |
gr.Markdown("---")
|
| 243 |
|
| 244 |
with gr.Row():
|
| 245 |
gr.Markdown("### GGUF Options")
|
| 246 |
with gr.Column():
|
| 247 |
-
|
| 248 |
-
|
| 249 |
-
|
| 250 |
with gr.Column():
|
| 251 |
-
|
| 252 |
-
|
| 253 |
gr.Markdown("---")
|
| 254 |
|
| 255 |
with gr.Row():
|
|
@@ -258,7 +297,6 @@ with gr.Blocks() as demo:
|
|
| 258 |
with gr.Column():
|
| 259 |
hub_model_name = gr.Textbox(label="Hub Model Name", value=f"username/model_name", interactive=True)
|
| 260 |
hub_token = gr.Textbox(label="Hub Token", interactive=True, type="password")
|
| 261 |
-
ollama_pub_key = gr.Button("HuggingFace Access Token")
|
| 262 |
gr.Markdown("---")
|
| 263 |
|
| 264 |
with gr.Row():
|
|
@@ -270,23 +308,21 @@ with gr.Blocks() as demo:
|
|
| 270 |
ollama_model_name = gr.Textbox(label="Ollama Model Name", value="user/model_name")
|
| 271 |
ollama_pub_key = gr.Button("Ollama Pub Key")
|
| 272 |
gr.Markdown("---")
|
| 273 |
-
|
|
|
|
| 274 |
|
| 275 |
with gr.Tab("Inference"):
|
| 276 |
with gr.Row():
|
| 277 |
-
gr.Textbox(label="Input Text", lines=4, value="""\
|
| 278 |
Continue the fibonnaci sequence.
|
| 279 |
# instruction
|
| 280 |
1, 1, 2, 3, 5, 8
|
| 281 |
# input
|
| 282 |
""", interactive=True)
|
| 283 |
-
gr.Textbox(label="Output Text", lines=4, value=""
|
| 284 |
-
""", interactive=False)
|
| 285 |
-
|
| 286 |
-
inference_button = gr.Button("Inference", visible=False, interactive=False)
|
| 287 |
-
# Output
|
| 288 |
|
| 289 |
-
|
|
|
|
| 290 |
load_btn.click(load_model, inputs=[initial_model_name, load_in_4bit, max_sequence_length], outputs=[output, load_btn, train_btn, initial_model_name, load_in_4bit, max_sequence_length])
|
| 291 |
|
| 292 |
demo.launch()
|
|
|
|
| 92 |
dataset = dataset.map(lambda examples: formatting_prompts_func(examples, data_template), batched=True)
|
| 93 |
return f"Data loaded {len(dataset)} records loaded.", gr.update(visible=True, interactive=True), gr.update(visible=True, interactive=True)
|
| 94 |
|
| 95 |
+
def inference(prompt, input_text):
|
| 96 |
+
FastLanguageModel.for_inference(model) # Enable native 2x faster inference
|
| 97 |
+
inputs = tokenizer(
|
| 98 |
+
[
|
| 99 |
+
prompt.format(
|
| 100 |
+
"Continue the fibonnaci sequence.", # instruction
|
| 101 |
+
"1, 1, 2, 3, 5, 8", # input
|
| 102 |
+
"", # output - leave this blank for generation!
|
| 103 |
+
)
|
| 104 |
+
], return_tensors = "pt").to("cuda")
|
| 105 |
+
|
| 106 |
+
outputs = model.generate(**inputs, max_new_tokens = 64, use_cache = True)
|
| 107 |
+
result = tokenizer.batch_decode(outputs)
|
| 108 |
+
return result[0], gr.update(visible=True, interactive=True)
|
| 109 |
|
| 110 |
|
| 111 |
async def train_model(model_name: str, lora_r: int, lora_alpha: int, lora_dropout: float, per_device_train_batch_size: int, warmup_steps: int, max_steps: int,
|
|
|
|
| 156 |
trainer.train()
|
| 157 |
return "Model training",gr.update(visible=True, interactive=False), gr.update(visible=True, interactive=True), gr.update(interactive=True)
|
| 158 |
|
| 159 |
+
def save_model(model_name, hub_model_name, hub_token, gguf_16bit, gguf_8bit, gguf_4bit, gguf_custom, gguf_custom_value, merge_16bit, merge_4bit, just_lora, push_to_hub):
|
| 160 |
+
global model, tokenizer
|
| 161 |
+
if gguf_custom:
|
| 162 |
+
gguf_custom_value = gguf_custom_value
|
| 163 |
+
else:
|
| 164 |
+
gguf_custom_value = None
|
| 165 |
+
|
| 166 |
+
if gguf_16bit:
|
| 167 |
+
gguf = "f16"
|
| 168 |
+
elif gguf_8bit:
|
| 169 |
+
gguf = "Q8_0"
|
| 170 |
+
elif gguf_4bit:
|
| 171 |
+
gguf = "q4_k_m"
|
| 172 |
+
else:
|
| 173 |
+
gguf = None
|
| 174 |
+
|
| 175 |
+
if merge_16bit:
|
| 176 |
+
merge = "16bit"
|
| 177 |
+
elif merge_4bit:
|
| 178 |
+
merge = "4bit"
|
| 179 |
+
elif just_lora:
|
| 180 |
+
merge = "lora"
|
| 181 |
+
else:
|
| 182 |
+
merge = None
|
| 183 |
+
|
| 184 |
+
#model.push_to_hub_gguf("hf/model", tokenizer, quantization_method = "f16", token = "")
|
| 185 |
+
if push_to_hub:
|
| 186 |
+
model.push_to_hub_gguf(hub_model_name, tokenizer, quantization_method=gguf, token=hub_token)
|
| 187 |
+
return "Model saved", gr.update(visible=True, interactive=True)
|
| 188 |
|
| 189 |
# Create the Gradio interface
|
| 190 |
with gr.Blocks() as demo:
|
|
|
|
| 210 |
dataset_name = gr.Textbox(label="Dataset Name", value="yahma/alpaca-cleaned")
|
| 211 |
data_template_style = gr.Dropdown(label="Template", choices=["alpaca","custom"], value="alpaca", allow_custom_value=True)
|
| 212 |
with gr.Row():
|
| 213 |
+
data_template = gr.TextArea(label="Data Template", value="""Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
|
| 214 |
|
| 215 |
### Instruction:
|
| 216 |
{}
|
|
|
|
| 223 |
gr.Markdown("---")
|
| 224 |
output_load_data = gr.Textbox(label="Data Load Status", value="Data not loaded", interactive=False)
|
| 225 |
load_data_btn = gr.Button("Load Dataset", interactive=True)
|
| 226 |
+
load_data_btn.click(load_data, inputs=[dataset_name, data_template_style, data_template], outputs=[output_load_data, load_data_btn])
|
| 227 |
|
| 228 |
with gr.Tab("Fine-Tuning"):
|
| 229 |
gr.Markdown("""### Fine-Tuned Model Parameters""")
|
|
|
|
| 277 |
with gr.Column():
|
| 278 |
merge_16bit = gr.Checkbox(label="Merge to 16bit", value=False, interactive=True)
|
| 279 |
merge_4bit = gr.Checkbox(label="Merge to 4bit", value=False, interactive=True)
|
| 280 |
+
just_lora = gr.Checkbox(label="Just LoRA Adapter", value=False, interactive=True)
|
| 281 |
gr.Markdown("---")
|
| 282 |
|
| 283 |
with gr.Row():
|
| 284 |
gr.Markdown("### GGUF Options")
|
| 285 |
with gr.Column():
|
| 286 |
+
gguf_16bit = gr.Checkbox(label="Quantize to f16", value=False, interactive=True)
|
| 287 |
+
gguf_8bit = gr.Checkbox(label="Quantize to 8bit (Q8_0)", value=False, interactive=True)
|
| 288 |
+
gguf_4bit = gr.Checkbox(label="Quantize to 4bit (q4_k_m)", value=False, interactive=True)
|
| 289 |
with gr.Column():
|
| 290 |
+
gguf_custom = gr.Checkbox(label="Custom", value=False, interactive=True)
|
| 291 |
+
gguf_custom_value = gr.Textbox(label="", value="Q5_K", interactive=True)
|
| 292 |
gr.Markdown("---")
|
| 293 |
|
| 294 |
with gr.Row():
|
|
|
|
| 297 |
with gr.Column():
|
| 298 |
hub_model_name = gr.Textbox(label="Hub Model Name", value=f"username/model_name", interactive=True)
|
| 299 |
hub_token = gr.Textbox(label="Hub Token", interactive=True, type="password")
|
|
|
|
| 300 |
gr.Markdown("---")
|
| 301 |
|
| 302 |
with gr.Row():
|
|
|
|
| 308 |
ollama_model_name = gr.Textbox(label="Ollama Model Name", value="user/model_name")
|
| 309 |
ollama_pub_key = gr.Button("Ollama Pub Key")
|
| 310 |
gr.Markdown("---")
|
| 311 |
+
save_button = gr.Button("Save Model", visible=True, interactive=True)
|
| 312 |
+
save_button.click(save_model, inputs=[model_name, hub_model_name, hub_token, gguf_16bit, gguf_8bit, gguf_4bit, gguf_custom, gguf_custom_value, merge_16bit, merge_4bit, just_lora, push_to_hub], outputs=[save_button])
|
| 313 |
|
| 314 |
with gr.Tab("Inference"):
|
| 315 |
with gr.Row():
|
| 316 |
+
input_text = gr.Textbox(label="Input Text", lines=4, value="""\
|
| 317 |
Continue the fibonnaci sequence.
|
| 318 |
# instruction
|
| 319 |
1, 1, 2, 3, 5, 8
|
| 320 |
# input
|
| 321 |
""", interactive=True)
|
| 322 |
+
output_text = gr.Textbox(label="Output Text", lines=4, value="", interactive=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 323 |
|
| 324 |
+
inference_button = gr.Button("Inference", visible=True, interactive=True)
|
| 325 |
+
inference_button.click(inference, inputs=[data_template, input_text], outputs=[output_text, inference_button])
|
| 326 |
load_btn.click(load_model, inputs=[initial_model_name, load_in_4bit, max_sequence_length], outputs=[output, load_btn, train_btn, initial_model_name, load_in_4bit, max_sequence_length])
|
| 327 |
|
| 328 |
demo.launch()
|