zethos
zethos
RRunPod
Created by zethos on 1/30/2025 in #⛅|pods
Cannot find any model weights with `/models/huggingface-cache/hub/models...`
Hi, I made a docker image using the "STEP-2" mentioned in Readme file. I created an template with docker image with below environment variables: MODEL_NAME="migtissera/Tess-3-Mistral-Large-2-123B" MAX_MODEL_LEN=65536 TENSOR_PARALLEL_SIZE=8 GPU_MEMORY_UTILIZATION=0.92 ENABLE_CHUNKED_PREFILL=1 NCCL_P2P_DISABLE=1 OMP_NUM_THREADS=1 ENFORCE_EAGER=1 The docker image: snbhanja/tess3mistrallarge128b:latest I tried to deploy this into a serverless with 8 48GB GPU. I get the below error but I didn't get this error when the very first time it is deployed: RuntimeError: Cannot find any model weights with `/models/huggingface-cache/hub/models--migtissera--Tess-3-Mistral-Large-2-123B/snapshots/8047f7cc9615909650b6a4ae5d13719d3e11594d Even if i delete the serveless endpoint and try to make one using this, it gives same error. Full log: https://github.com/user-attachments/files/18603761/logs.11.txt
2 replies
RRunPod
Created by zethos on 1/29/2025 in #⚡|serverless
Need help in fixing long running deployments in serverless vLLM
No description
18 replies
RRunPod
Created by zethos on 1/13/2024 in #⚡|serverless
#How to upload a file using a upload api in gpu serverless?
This is my current code, there is a separate fastapi running.
automatic_session = requests.Session()
retries = Retry(total=10, backoff_factor=0.1, status_forcelist=[502, 503, 504])
automatic_session.mount('http://', HTTPAdapter(max_retries=retries))

def run_inference(params):
config = {
"baseurl": "http://127.0.0.1:8080",
"api": {
"health_check": ("GET", "/health_check"),
"upload": ("POST", "/upload"),
},
"timeout": 600
}

api_name = params["api_name"]

if api_name in config["api"]:
api_config = config["api"][api_name]
else:
raise Exception("Method '%s' not yet implemented")

api_verb = api_config[0]
api_path = api_config[1]

response = {}

if api_verb == "GET":
response = automatic_session.get(
url='%s%s' % (config["baseurl"], api_path),
timeout=config["timeout"])

elif api_verb == "POST":
if "upload" in api_path:
print("Inside upload rp")
# For "upload" endpoints, use form data
files = {'file': ('filename', params['file'].read())} # Assuming the file is passed in params
response = automatic_session.post(
url='%s%s' % (config["baseurl"], api_path),
files=files,
timeout=config["timeout"])
else:
# For other POST requests, use JSON
response = automatic_session.post(
url='%s%s' % (config["baseurl"], api_path),
json=params,
timeout=config["timeout"])

return response.json()
def handler(event):
json = run_inference(event["input"])
return json

if __name__ == "__main__":
runpod.serverless.start({"handler": handler})
automatic_session = requests.Session()
retries = Retry(total=10, backoff_factor=0.1, status_forcelist=[502, 503, 504])
automatic_session.mount('http://', HTTPAdapter(max_retries=retries))

def run_inference(params):
config = {
"baseurl": "http://127.0.0.1:8080",
"api": {
"health_check": ("GET", "/health_check"),
"upload": ("POST", "/upload"),
},
"timeout": 600
}

api_name = params["api_name"]

if api_name in config["api"]:
api_config = config["api"][api_name]
else:
raise Exception("Method '%s' not yet implemented")

api_verb = api_config[0]
api_path = api_config[1]

response = {}

if api_verb == "GET":
response = automatic_session.get(
url='%s%s' % (config["baseurl"], api_path),
timeout=config["timeout"])

elif api_verb == "POST":
if "upload" in api_path:
print("Inside upload rp")
# For "upload" endpoints, use form data
files = {'file': ('filename', params['file'].read())} # Assuming the file is passed in params
response = automatic_session.post(
url='%s%s' % (config["baseurl"], api_path),
files=files,
timeout=config["timeout"])
else:
# For other POST requests, use JSON
response = automatic_session.post(
url='%s%s' % (config["baseurl"], api_path),
json=params,
timeout=config["timeout"])

return response.json()
def handler(event):
json = run_inference(event["input"])
return json

if __name__ == "__main__":
runpod.serverless.start({"handler": handler})
#⚡|serverless
4 replies