Spaces:
Running
on
Zero
Running
on
Zero
Commit
·
a7d545f
1
Parent(s):
5e5d5d1
add 286 mix6 table
Browse files- app_dialogue.py +11 -4
app_dialogue.py
CHANGED
|
@@ -25,20 +25,27 @@ subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENT
|
|
| 25 |
|
| 26 |
DEVICE = torch.device("cuda")
|
| 27 |
MODELS = {
|
| 28 |
-
"
|
| 29 |
"HuggingFaceM4/idefics2",
|
| 30 |
trust_remote_code=True,
|
| 31 |
torch_dtype=torch.bfloat16,
|
| 32 |
token=os.environ["HF_AUTH_TOKEN"],
|
| 33 |
revision="1e05755c1c5cb2077a0f60b83ea1368c22a17282",
|
| 34 |
).to(DEVICE),
|
| 35 |
-
"
|
| 36 |
"HuggingFaceM4/idefics2",
|
| 37 |
trust_remote_code=True,
|
| 38 |
torch_dtype=torch.bfloat16,
|
| 39 |
token=os.environ["HF_AUTH_TOKEN"],
|
| 40 |
revision="5cd3c3a3eb5e0ea664f5ac09e73c9ef42da93a86",
|
| 41 |
).to(DEVICE),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 42 |
}
|
| 43 |
PROCESSOR = AutoProcessor.from_pretrained(
|
| 44 |
"HuggingFaceM4/idefics2",
|
|
@@ -331,7 +338,7 @@ with gr.Blocks(title="IDEFICS Playground", theme=gr.themes.Base()) as demo:
|
|
| 331 |
with gr.Row(elem_id="model_selector_row"):
|
| 332 |
model_selector = gr.Dropdown(
|
| 333 |
choices=MODELS.keys(),
|
| 334 |
-
value="
|
| 335 |
interactive=True,
|
| 336 |
show_label=False,
|
| 337 |
container=False,
|
|
@@ -520,7 +527,7 @@ with gr.Blocks(title="IDEFICS Playground", theme=gr.themes.Base()) as demo:
|
|
| 520 |
Same as `model_inference` but in greedy mode and with the 80b-instruct.
|
| 521 |
Specifically for pre-computing the default examples.
|
| 522 |
"""
|
| 523 |
-
model_selector = "
|
| 524 |
user_prompt_str = message
|
| 525 |
chat_history = []
|
| 526 |
max_new_tokens = 512
|
|
|
|
| 25 |
|
| 26 |
DEVICE = torch.device("cuda")
|
| 27 |
MODELS = {
|
| 28 |
+
"284 - neftune - opt 18'500": AutoModelForCausalLM.from_pretrained(
|
| 29 |
"HuggingFaceM4/idefics2",
|
| 30 |
trust_remote_code=True,
|
| 31 |
torch_dtype=torch.bfloat16,
|
| 32 |
token=os.environ["HF_AUTH_TOKEN"],
|
| 33 |
revision="1e05755c1c5cb2077a0f60b83ea1368c22a17282",
|
| 34 |
).to(DEVICE),
|
| 35 |
+
"279bis - baseline - opt 18'500": AutoModelForCausalLM.from_pretrained(
|
| 36 |
"HuggingFaceM4/idefics2",
|
| 37 |
trust_remote_code=True,
|
| 38 |
torch_dtype=torch.bfloat16,
|
| 39 |
token=os.environ["HF_AUTH_TOKEN"],
|
| 40 |
revision="5cd3c3a3eb5e0ea664f5ac09e73c9ef42da93a86",
|
| 41 |
).to(DEVICE),
|
| 42 |
+
"286 - mix6 tables - opt 20'000": AutoModelForCausalLM.from_pretrained(
|
| 43 |
+
"HuggingFaceM4/idefics2",
|
| 44 |
+
trust_remote_code=True,
|
| 45 |
+
torch_dtype=torch.bfloat16,
|
| 46 |
+
token=os.environ["HF_AUTH_TOKEN"],
|
| 47 |
+
revision="b473d49caa964991b40b79fe7cb27d51d4d023f6",
|
| 48 |
+
).to(DEVICE),
|
| 49 |
}
|
| 50 |
PROCESSOR = AutoProcessor.from_pretrained(
|
| 51 |
"HuggingFaceM4/idefics2",
|
|
|
|
| 338 |
with gr.Row(elem_id="model_selector_row"):
|
| 339 |
model_selector = gr.Dropdown(
|
| 340 |
choices=MODELS.keys(),
|
| 341 |
+
value="284 - neftune - opt 18'500",
|
| 342 |
interactive=True,
|
| 343 |
show_label=False,
|
| 344 |
container=False,
|
|
|
|
| 527 |
Same as `model_inference` but in greedy mode and with the 80b-instruct.
|
| 528 |
Specifically for pre-computing the default examples.
|
| 529 |
"""
|
| 530 |
+
model_selector = "284 - neftune - opt 18'500"
|
| 531 |
user_prompt_str = message
|
| 532 |
chat_history = []
|
| 533 |
max_new_tokens = 512
|