Update server.py
Browse files
server.py
CHANGED
|
@@ -320,34 +320,57 @@ def generate_report(
|
|
| 320 |
|
| 321 |
|
| 322 |
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 323 |
-
# LOAD MODELS FROM HUGGINGFACE
|
| 324 |
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 325 |
print("\n" + "="*80)
|
| 326 |
-
print("
|
| 327 |
print("="*80)
|
| 328 |
|
|
|
|
|
|
|
|
|
|
| 329 |
# Download model files from Hugging Face
|
| 330 |
try:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 331 |
SFT_MODEL_PATH = hf_hub_download(
|
| 332 |
-
repo_id=
|
| 333 |
filename="best_model.pt"
|
| 334 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 335 |
PPO_MODEL_PATH = hf_hub_download(
|
| 336 |
-
repo_id=
|
| 337 |
filename="rlhf_model.pt"
|
| 338 |
)
|
| 339 |
-
print(f"β
|
| 340 |
-
|
|
|
|
|
|
|
| 341 |
except Exception as e:
|
| 342 |
-
print(f"β Error downloading models: {e}")
|
| 343 |
-
|
| 344 |
-
|
| 345 |
-
|
| 346 |
-
print(
|
|
|
|
| 347 |
|
| 348 |
# Load both models - EXACTLY as Colab SECTION 8
|
| 349 |
print("\n" + "="*80)
|
| 350 |
-
print("LOADING MODELS")
|
| 351 |
print("="*80)
|
| 352 |
|
| 353 |
sft_model = load_model_from_checkpoint(
|
|
@@ -362,12 +385,17 @@ ppo_model = load_model_from_checkpoint(
|
|
| 362 |
CONFIG
|
| 363 |
)
|
| 364 |
|
| 365 |
-
print("\n
|
|
|
|
| 366 |
|
| 367 |
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 368 |
# FASTAPI APP
|
| 369 |
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 370 |
-
app = FastAPI(
|
|
|
|
|
|
|
|
|
|
|
|
|
| 371 |
|
| 372 |
app.add_middleware(
|
| 373 |
CORSMiddleware,
|
|
@@ -377,14 +405,40 @@ app.add_middleware(
|
|
| 377 |
)
|
| 378 |
|
| 379 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 380 |
@app.get("/health")
|
| 381 |
def health():
|
| 382 |
return {
|
| 383 |
"status": "ok",
|
| 384 |
"device": str(device),
|
| 385 |
"cuda_available": torch.cuda.is_available(),
|
| 386 |
-
"models_loaded":
|
| 387 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 388 |
}
|
| 389 |
|
| 390 |
|
|
@@ -392,6 +446,8 @@ def health():
|
|
| 392 |
async def sft_inference(file: UploadFile = File(...)):
|
| 393 |
"""
|
| 394 |
SFT model inference - Uses EXACT generate_report() function from Colab SECTION 9
|
|
|
|
|
|
|
| 395 |
"""
|
| 396 |
try:
|
| 397 |
# Save uploaded file temporarily
|
|
@@ -410,18 +466,24 @@ async def sft_inference(file: UploadFile = File(...)):
|
|
| 410 |
return {
|
| 411 |
"report": report,
|
| 412 |
"model": "SFT",
|
| 413 |
-
"
|
|
|
|
| 414 |
}
|
| 415 |
|
| 416 |
except Exception as e:
|
| 417 |
traceback.print_exc()
|
| 418 |
-
return {
|
|
|
|
|
|
|
|
|
|
| 419 |
|
| 420 |
|
| 421 |
@app.post("/ppo")
|
| 422 |
async def ppo_inference(file: UploadFile = File(...)):
|
| 423 |
"""
|
| 424 |
PPO model inference - Uses EXACT generate_report() function from Colab SECTION 9
|
|
|
|
|
|
|
| 425 |
"""
|
| 426 |
try:
|
| 427 |
# Save uploaded file temporarily
|
|
@@ -440,12 +502,16 @@ async def ppo_inference(file: UploadFile = File(...)):
|
|
| 440 |
return {
|
| 441 |
"report": report,
|
| 442 |
"model": "PPO",
|
| 443 |
-
"
|
|
|
|
| 444 |
}
|
| 445 |
|
| 446 |
except Exception as e:
|
| 447 |
traceback.print_exc()
|
| 448 |
-
return {
|
|
|
|
|
|
|
|
|
|
| 449 |
|
| 450 |
|
| 451 |
@app.post("/compare")
|
|
@@ -453,6 +519,8 @@ async def compare_models(file: UploadFile = File(...)):
|
|
| 453 |
"""
|
| 454 |
Generate reports from both models for comparison
|
| 455 |
Uses EXACT generate_report() function from Colab
|
|
|
|
|
|
|
| 456 |
"""
|
| 457 |
try:
|
| 458 |
# Save uploaded file temporarily
|
|
@@ -473,8 +541,11 @@ async def compare_models(file: UploadFile = File(...)):
|
|
| 473 |
return {
|
| 474 |
"sft_report": sft_report,
|
| 475 |
"ppo_report": ppo_report,
|
| 476 |
-
"
|
| 477 |
-
|
|
|
|
|
|
|
|
|
|
| 478 |
}
|
| 479 |
|
| 480 |
except Exception as e:
|
|
@@ -485,43 +556,45 @@ async def compare_models(file: UploadFile = File(...)):
|
|
| 485 |
}
|
| 486 |
|
| 487 |
|
| 488 |
-
@app.get("/
|
| 489 |
-
def
|
| 490 |
"""
|
| 491 |
-
|
| 492 |
"""
|
| 493 |
return {
|
| 494 |
-
"
|
| 495 |
-
"
|
| 496 |
-
"
|
| 497 |
-
"
|
| 498 |
-
|
| 499 |
-
|
| 500 |
-
|
| 501 |
-
|
| 502 |
-
|
| 503 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 504 |
},
|
| 505 |
-
"
|
| 506 |
-
|
| 507 |
-
"
|
| 508 |
-
"
|
| 509 |
-
"normalize_std": [0.229, 0.224, 0.225]
|
| 510 |
},
|
| 511 |
-
"
|
| 512 |
"max_length": CONFIG['max_length'],
|
| 513 |
"num_beams": CONFIG['num_beams'],
|
| 514 |
-
"
|
| 515 |
-
"no_extra_penalties": "β Exactly as Colab"
|
| 516 |
-
},
|
| 517 |
-
"inference_method": "generate_report() from Colab SECTION 9",
|
| 518 |
-
"models_loaded": {
|
| 519 |
-
"sft": sft_model is not None,
|
| 520 |
-
"ppo": ppo_model is not None
|
| 521 |
-
},
|
| 522 |
-
"model_state": {
|
| 523 |
-
"sft_eval_mode": not sft_model.training if sft_model else None,
|
| 524 |
-
"ppo_eval_mode": not ppo_model.training if ppo_model else None
|
| 525 |
}
|
| 526 |
}
|
| 527 |
|
|
@@ -538,15 +611,19 @@ else:
|
|
| 538 |
print("β οΈ Build directory not found, serving API only")
|
| 539 |
|
| 540 |
print("\n" + "="*80)
|
| 541 |
-
print("SERVER READY
|
| 542 |
print("="*80)
|
| 543 |
-
print("
|
| 544 |
-
print("
|
| 545 |
-
print(" β
|
| 546 |
-
print(" β
|
| 547 |
-
print("
|
| 548 |
-
print("
|
| 549 |
-
print("
|
|
|
|
|
|
|
|
|
|
|
|
|
| 550 |
print("="*80)
|
| 551 |
|
| 552 |
if __name__ == "__main__":
|
|
|
|
| 320 |
|
| 321 |
|
| 322 |
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 323 |
+
# LOAD MODELS FROM HUGGINGFACE - Shree2604/BioStack
|
| 324 |
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 325 |
print("\n" + "="*80)
|
| 326 |
+
print("DOWNLOADING MODELS FROM HUGGINGFACE")
|
| 327 |
print("="*80)
|
| 328 |
|
| 329 |
+
# Hugging Face repository
|
| 330 |
+
HF_REPO = "Shree2604/BioStack"
|
| 331 |
+
|
| 332 |
# Download model files from Hugging Face
|
| 333 |
try:
|
| 334 |
+
print(f"π¦ Downloading from repository: {HF_REPO}")
|
| 335 |
+
print("This may take a few minutes on first run...\n")
|
| 336 |
+
|
| 337 |
+
# Download SFT model
|
| 338 |
+
print("1οΈβ£ Downloading SFT model (best_model.pt)...")
|
| 339 |
SFT_MODEL_PATH = hf_hub_download(
|
| 340 |
+
repo_id=HF_REPO,
|
| 341 |
filename="best_model.pt"
|
| 342 |
)
|
| 343 |
+
print(f" β SFT model downloaded: {SFT_MODEL_PATH}")
|
| 344 |
+
|
| 345 |
+
# Download Reward model
|
| 346 |
+
print("\n2οΈβ£ Downloading Reward model (reward_model.pt)...")
|
| 347 |
+
REWARD_MODEL_PATH = hf_hub_download(
|
| 348 |
+
repo_id=HF_REPO,
|
| 349 |
+
filename="reward_model.pt"
|
| 350 |
+
)
|
| 351 |
+
print(f" β Reward model downloaded: {REWARD_MODEL_PATH}")
|
| 352 |
+
|
| 353 |
+
# Download PPO model
|
| 354 |
+
print("\n3οΈβ£ Downloading PPO model (rlhf_model.pt)...")
|
| 355 |
PPO_MODEL_PATH = hf_hub_download(
|
| 356 |
+
repo_id=HF_REPO,
|
| 357 |
filename="rlhf_model.pt"
|
| 358 |
)
|
| 359 |
+
print(f" β PPO model downloaded: {PPO_MODEL_PATH}")
|
| 360 |
+
|
| 361 |
+
print("\nβ
All models downloaded successfully!")
|
| 362 |
+
|
| 363 |
except Exception as e:
|
| 364 |
+
print(f"\nβ Error downloading models from Hugging Face: {e}")
|
| 365 |
+
print("Please check:")
|
| 366 |
+
print(f" - Repository exists: https://huggingface.co/{HF_REPO}")
|
| 367 |
+
print(" - Files exist: best_model.pt, reward_model.pt, rlhf_model.pt")
|
| 368 |
+
print(" - You have internet connection")
|
| 369 |
+
raise
|
| 370 |
|
| 371 |
# Load both models - EXACTLY as Colab SECTION 8
|
| 372 |
print("\n" + "="*80)
|
| 373 |
+
print("LOADING MODELS INTO MEMORY")
|
| 374 |
print("="*80)
|
| 375 |
|
| 376 |
sft_model = load_model_from_checkpoint(
|
|
|
|
| 385 |
CONFIG
|
| 386 |
)
|
| 387 |
|
| 388 |
+
print("\nβ
Both models loaded successfully!")
|
| 389 |
+
print("="*80)
|
| 390 |
|
| 391 |
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 392 |
# FASTAPI APP
|
| 393 |
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 394 |
+
app = FastAPI(
|
| 395 |
+
title="BioStack Medical Report Generation",
|
| 396 |
+
description="Medical X-ray report generation using SFT and PPO models from Shree2604/BioStack",
|
| 397 |
+
version="1.0.0"
|
| 398 |
+
)
|
| 399 |
|
| 400 |
app.add_middleware(
|
| 401 |
CORSMiddleware,
|
|
|
|
| 405 |
)
|
| 406 |
|
| 407 |
|
| 408 |
+
@app.get("/")
|
| 409 |
+
def root():
|
| 410 |
+
return {
|
| 411 |
+
"message": "BioStack Medical Report Generation API",
|
| 412 |
+
"repository": "Shree2604/BioStack",
|
| 413 |
+
"models": {
|
| 414 |
+
"sft": "best_model.pt",
|
| 415 |
+
"ppo": "rlhf_model.pt",
|
| 416 |
+
"reward": "reward_model.pt"
|
| 417 |
+
},
|
| 418 |
+
"endpoints": {
|
| 419 |
+
"health": "GET /health - Check API status",
|
| 420 |
+
"sft": "POST /sft - Generate report using SFT model",
|
| 421 |
+
"ppo": "POST /ppo - Generate report using PPO model",
|
| 422 |
+
"compare": "POST /compare - Compare both models"
|
| 423 |
+
}
|
| 424 |
+
}
|
| 425 |
+
|
| 426 |
+
|
| 427 |
@app.get("/health")
|
| 428 |
def health():
|
| 429 |
return {
|
| 430 |
"status": "ok",
|
| 431 |
"device": str(device),
|
| 432 |
"cuda_available": torch.cuda.is_available(),
|
| 433 |
+
"models_loaded": {
|
| 434 |
+
"sft": sft_model is not None,
|
| 435 |
+
"ppo": ppo_model is not None
|
| 436 |
+
},
|
| 437 |
+
"repository": HF_REPO,
|
| 438 |
+
"model_files": {
|
| 439 |
+
"sft": os.path.basename(SFT_MODEL_PATH),
|
| 440 |
+
"ppo": os.path.basename(PPO_MODEL_PATH)
|
| 441 |
+
}
|
| 442 |
}
|
| 443 |
|
| 444 |
|
|
|
|
| 446 |
async def sft_inference(file: UploadFile = File(...)):
|
| 447 |
"""
|
| 448 |
SFT model inference - Uses EXACT generate_report() function from Colab SECTION 9
|
| 449 |
+
|
| 450 |
+
Model: best_model.pt from Shree2604/BioStack
|
| 451 |
"""
|
| 452 |
try:
|
| 453 |
# Save uploaded file temporarily
|
|
|
|
| 466 |
return {
|
| 467 |
"report": report,
|
| 468 |
"model": "SFT",
|
| 469 |
+
"source": "best_model.pt",
|
| 470 |
+
"repository": HF_REPO
|
| 471 |
}
|
| 472 |
|
| 473 |
except Exception as e:
|
| 474 |
traceback.print_exc()
|
| 475 |
+
return {
|
| 476 |
+
"report": f"ERROR: {str(e)}",
|
| 477 |
+
"model": "SFT"
|
| 478 |
+
}
|
| 479 |
|
| 480 |
|
| 481 |
@app.post("/ppo")
|
| 482 |
async def ppo_inference(file: UploadFile = File(...)):
|
| 483 |
"""
|
| 484 |
PPO model inference - Uses EXACT generate_report() function from Colab SECTION 9
|
| 485 |
+
|
| 486 |
+
Model: rlhf_model.pt from Shree2604/BioStack
|
| 487 |
"""
|
| 488 |
try:
|
| 489 |
# Save uploaded file temporarily
|
|
|
|
| 502 |
return {
|
| 503 |
"report": report,
|
| 504 |
"model": "PPO",
|
| 505 |
+
"source": "rlhf_model.pt",
|
| 506 |
+
"repository": HF_REPO
|
| 507 |
}
|
| 508 |
|
| 509 |
except Exception as e:
|
| 510 |
traceback.print_exc()
|
| 511 |
+
return {
|
| 512 |
+
"report": f"ERROR: {str(e)}",
|
| 513 |
+
"model": "PPO"
|
| 514 |
+
}
|
| 515 |
|
| 516 |
|
| 517 |
@app.post("/compare")
|
|
|
|
| 519 |
"""
|
| 520 |
Generate reports from both models for comparison
|
| 521 |
Uses EXACT generate_report() function from Colab
|
| 522 |
+
|
| 523 |
+
Models: best_model.pt and rlhf_model.pt from Shree2604/BioStack
|
| 524 |
"""
|
| 525 |
try:
|
| 526 |
# Save uploaded file temporarily
|
|
|
|
| 541 |
return {
|
| 542 |
"sft_report": sft_report,
|
| 543 |
"ppo_report": ppo_report,
|
| 544 |
+
"models": {
|
| 545 |
+
"sft": "best_model.pt",
|
| 546 |
+
"ppo": "rlhf_model.pt"
|
| 547 |
+
},
|
| 548 |
+
"repository": HF_REPO
|
| 549 |
}
|
| 550 |
|
| 551 |
except Exception as e:
|
|
|
|
| 556 |
}
|
| 557 |
|
| 558 |
|
| 559 |
+
@app.get("/model_info")
|
| 560 |
+
def model_info():
|
| 561 |
"""
|
| 562 |
+
Get detailed information about loaded models
|
| 563 |
"""
|
| 564 |
return {
|
| 565 |
+
"repository": HF_REPO,
|
| 566 |
+
"repository_url": f"https://huggingface.co/{HF_REPO}",
|
| 567 |
+
"models": {
|
| 568 |
+
"sft": {
|
| 569 |
+
"filename": "best_model.pt",
|
| 570 |
+
"url": f"https://huggingface.co/{HF_REPO}/blob/main/best_model.pt",
|
| 571 |
+
"local_path": SFT_MODEL_PATH,
|
| 572 |
+
"loaded": sft_model is not None,
|
| 573 |
+
"in_eval_mode": not sft_model.training if sft_model else None
|
| 574 |
+
},
|
| 575 |
+
"ppo": {
|
| 576 |
+
"filename": "rlhf_model.pt",
|
| 577 |
+
"url": f"https://huggingface.co/{HF_REPO}/blob/main/rlhf_model.pt",
|
| 578 |
+
"local_path": PPO_MODEL_PATH,
|
| 579 |
+
"loaded": ppo_model is not None,
|
| 580 |
+
"in_eval_mode": not ppo_model.training if ppo_model else None
|
| 581 |
+
},
|
| 582 |
+
"reward": {
|
| 583 |
+
"filename": "reward_model.pt",
|
| 584 |
+
"url": f"https://huggingface.co/{HF_REPO}/blob/main/reward_model.pt",
|
| 585 |
+
"local_path": REWARD_MODEL_PATH,
|
| 586 |
+
"note": "Downloaded but not loaded in this API"
|
| 587 |
+
}
|
| 588 |
},
|
| 589 |
+
"architecture": {
|
| 590 |
+
"vision_encoder": CONFIG['coatnet_model'],
|
| 591 |
+
"text_model": CONFIG['t5_model'],
|
| 592 |
+
"image_embedding_dim": CONFIG['img_emb_dim']
|
|
|
|
| 593 |
},
|
| 594 |
+
"inference_config": {
|
| 595 |
"max_length": CONFIG['max_length'],
|
| 596 |
"num_beams": CONFIG['num_beams'],
|
| 597 |
+
"image_size": CONFIG['image_size']
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 598 |
}
|
| 599 |
}
|
| 600 |
|
|
|
|
| 611 |
print("β οΈ Build directory not found, serving API only")
|
| 612 |
|
| 613 |
print("\n" + "="*80)
|
| 614 |
+
print("π SERVER READY")
|
| 615 |
print("="*80)
|
| 616 |
+
print(f"Repository: {HF_REPO}")
|
| 617 |
+
print("Models loaded:")
|
| 618 |
+
print(f" β SFT: best_model.pt")
|
| 619 |
+
print(f" β PPO: rlhf_model.pt")
|
| 620 |
+
print("\nEndpoints:")
|
| 621 |
+
print(" GET / - API info")
|
| 622 |
+
print(" GET /health - Health check")
|
| 623 |
+
print(" GET /model_info - Model details")
|
| 624 |
+
print(" POST /sft - SFT inference")
|
| 625 |
+
print(" POST /ppo - PPO inference")
|
| 626 |
+
print(" POST /compare - Compare both models")
|
| 627 |
print("="*80)
|
| 628 |
|
| 629 |
if __name__ == "__main__":
|