Spaces:
Sleeping
Sleeping
Joash
commited on
Commit
·
1f37a6a
1
Parent(s):
93aa8dc
Add detailed logging and improve error handling in model manager
Browse files- Dockerfile +4 -2
- src/model_manager.py +22 -20
Dockerfile
CHANGED
@@ -26,6 +26,8 @@ ENV PORT=7860
|
|
26 |
ENV PATH="/home/user/.local/bin:${PATH}"
|
27 |
ENV HF_HOME=/home/user/.cache/huggingface
|
28 |
ENV TRANSFORMERS_CACHE=/home/user/.cache/huggingface
|
|
|
|
|
29 |
|
30 |
# Switch to non-root user
|
31 |
USER user
|
@@ -46,5 +48,5 @@ COPY --chown=user:user . .
|
|
46 |
# Expose port for Hugging Face Spaces
|
47 |
EXPOSE 7860
|
48 |
|
49 |
-
# Run the application
|
50 |
-
CMD ["python", "-m", "uvicorn", "src.api:app", "--host", "0.0.0.0", "--port", "7860"]
|
|
|
26 |
ENV PATH="/home/user/.local/bin:${PATH}"
|
27 |
ENV HF_HOME=/home/user/.cache/huggingface
|
28 |
ENV TRANSFORMERS_CACHE=/home/user/.cache/huggingface
|
29 |
+
# Set logging to stdout
|
30 |
+
ENV LOG_FILE=/dev/stdout
|
31 |
|
32 |
# Switch to non-root user
|
33 |
USER user
|
|
|
48 |
# Expose port for Hugging Face Spaces
|
49 |
EXPOSE 7860
|
50 |
|
51 |
+
# Run the application with logging
|
52 |
+
CMD ["python", "-u", "-m", "uvicorn", "src.api:app", "--host", "0.0.0.0", "--port", "7860", "--log-level", "debug"]
|
src/model_manager.py
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
import logging
|
2 |
-
from transformers import AutoTokenizer, AutoModelForCausalLM
|
3 |
import torch
|
4 |
from huggingface_hub import login
|
5 |
from .config import Config
|
@@ -16,7 +16,12 @@ class ModelManager:
|
|
16 |
# Login to Hugging Face Hub
|
17 |
if Config.HUGGING_FACE_TOKEN:
|
18 |
logger.info("Logging in to Hugging Face Hub")
|
19 |
-
|
|
|
|
|
|
|
|
|
|
|
20 |
|
21 |
# Initialize tokenizer and model
|
22 |
self._init_tokenizer()
|
@@ -37,7 +42,8 @@ class ModelManager:
|
|
37 |
'bos_token': '<s>'
|
38 |
}
|
39 |
self.tokenizer.add_special_tokens(special_tokens)
|
40 |
-
logger.info("Tokenizer loaded successfully
|
|
|
41 |
except Exception as e:
|
42 |
logger.error(f"Error loading tokenizer: {str(e)}")
|
43 |
raise
|
@@ -46,6 +52,7 @@ class ModelManager:
|
|
46 |
"""Initialize the model."""
|
47 |
try:
|
48 |
logger.info(f"Loading model: {self.model_name}")
|
|
|
49 |
|
50 |
# Load model with CPU configuration
|
51 |
self.model = AutoModelForCausalLM.from_pretrained(
|
@@ -57,7 +64,8 @@ class ModelManager:
|
|
57 |
)
|
58 |
# Resize embeddings to match tokenizer
|
59 |
self.model.resize_token_embeddings(len(self.tokenizer))
|
60 |
-
logger.info(
|
|
|
61 |
except Exception as e:
|
62 |
logger.error(f"Error loading model: {str(e)}")
|
63 |
raise
|
@@ -65,11 +73,16 @@ class ModelManager:
|
|
65 |
def generate_text(self, prompt: str, max_new_tokens: int = 1024) -> str:
|
66 |
"""Generate text from prompt."""
|
67 |
try:
|
|
|
|
|
|
|
68 |
# Encode the prompt
|
69 |
inputs = self.tokenizer(prompt, return_tensors="pt", truncation=True, max_length=2048)
|
70 |
inputs = {k: v.to(self.device) for k, v in inputs.items()}
|
|
|
71 |
|
72 |
# Generate response
|
|
|
73 |
with torch.no_grad():
|
74 |
outputs = self.model.generate(
|
75 |
**inputs,
|
@@ -80,27 +93,16 @@ class ModelManager:
|
|
80 |
pad_token_id=self.tokenizer.pad_token_id,
|
81 |
eos_token_id=self.tokenizer.eos_token_id,
|
82 |
)
|
83 |
-
|
84 |
# Decode and return the generated text
|
85 |
generated_text = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
|
86 |
-
# Extract only the generated part (remove the prompt)
|
87 |
response = generated_text[len(prompt):].strip()
|
88 |
|
|
|
|
|
89 |
return response
|
90 |
|
91 |
except Exception as e:
|
92 |
logger.error(f"Error generating text: {str(e)}")
|
93 |
-
|
94 |
-
|
95 |
-
- Model inference failed
|
96 |
-
|
97 |
-
- Improvements:
|
98 |
-
- Please try again
|
99 |
-
- Check model configuration
|
100 |
-
|
101 |
-
- Best Practices:
|
102 |
-
- Ensure proper model setup
|
103 |
-
- Verify token permissions
|
104 |
-
|
105 |
-
- Security:
|
106 |
-
- No immediate concerns"""
|
|
|
1 |
import logging
|
2 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer
|
3 |
import torch
|
4 |
from huggingface_hub import login
|
5 |
from .config import Config
|
|
|
16 |
# Login to Hugging Face Hub
|
17 |
if Config.HUGGING_FACE_TOKEN:
|
18 |
logger.info("Logging in to Hugging Face Hub")
|
19 |
+
try:
|
20 |
+
login(token=Config.HUGGING_FACE_TOKEN)
|
21 |
+
logger.info("Successfully logged in to Hugging Face Hub")
|
22 |
+
except Exception as e:
|
23 |
+
logger.error(f"Failed to login to Hugging Face Hub: {str(e)}")
|
24 |
+
raise
|
25 |
|
26 |
# Initialize tokenizer and model
|
27 |
self._init_tokenizer()
|
|
|
42 |
'bos_token': '<s>'
|
43 |
}
|
44 |
self.tokenizer.add_special_tokens(special_tokens)
|
45 |
+
logger.info("Tokenizer loaded successfully")
|
46 |
+
logger.debug(f"Tokenizer vocabulary size: {len(self.tokenizer)}")
|
47 |
except Exception as e:
|
48 |
logger.error(f"Error loading tokenizer: {str(e)}")
|
49 |
raise
|
|
|
52 |
"""Initialize the model."""
|
53 |
try:
|
54 |
logger.info(f"Loading model: {self.model_name}")
|
55 |
+
logger.info(f"Using device: {self.device}")
|
56 |
|
57 |
# Load model with CPU configuration
|
58 |
self.model = AutoModelForCausalLM.from_pretrained(
|
|
|
64 |
)
|
65 |
# Resize embeddings to match tokenizer
|
66 |
self.model.resize_token_embeddings(len(self.tokenizer))
|
67 |
+
logger.info("Model loaded successfully")
|
68 |
+
logger.debug(f"Model parameters: {sum(p.numel() for p in self.model.parameters())}")
|
69 |
except Exception as e:
|
70 |
logger.error(f"Error loading model: {str(e)}")
|
71 |
raise
|
|
|
73 |
def generate_text(self, prompt: str, max_new_tokens: int = 1024) -> str:
|
74 |
"""Generate text from prompt."""
|
75 |
try:
|
76 |
+
logger.info("Starting text generation")
|
77 |
+
logger.debug(f"Prompt length: {len(prompt)}")
|
78 |
+
|
79 |
# Encode the prompt
|
80 |
inputs = self.tokenizer(prompt, return_tensors="pt", truncation=True, max_length=2048)
|
81 |
inputs = {k: v.to(self.device) for k, v in inputs.items()}
|
82 |
+
logger.debug(f"Input tensor shape: {inputs['input_ids'].shape}")
|
83 |
|
84 |
# Generate response
|
85 |
+
logger.info("Generating response")
|
86 |
with torch.no_grad():
|
87 |
outputs = self.model.generate(
|
88 |
**inputs,
|
|
|
93 |
pad_token_id=self.tokenizer.pad_token_id,
|
94 |
eos_token_id=self.tokenizer.eos_token_id,
|
95 |
)
|
96 |
+
|
97 |
# Decode and return the generated text
|
98 |
generated_text = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
|
|
|
99 |
response = generated_text[len(prompt):].strip()
|
100 |
|
101 |
+
logger.info("Text generation completed")
|
102 |
+
logger.debug(f"Response length: {len(response)}")
|
103 |
return response
|
104 |
|
105 |
except Exception as e:
|
106 |
logger.error(f"Error generating text: {str(e)}")
|
107 |
+
logger.error(f"Error details: {type(e).__name__}")
|
108 |
+
raise
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|