mirror of
https://github.com/justUmen/Bjornulf_custom_nodes.git
synced 2026-03-21 20:52:11 -03:00
98 lines
4.6 KiB
Python
98 lines
4.6 KiB
Python
import ollama
|
|
from ollama import Client # pip install ollama
|
|
import logging
|
|
import hashlib
|
|
|
|
class ollamaLoader:
|
|
@classmethod
|
|
def get_available_models(cls):
|
|
try:
|
|
# First try with 127.0.0.1
|
|
client = Client(host="http://127.0.0.1:11434")
|
|
list_models = client.list() # Assuming list() is part of the Client class
|
|
return [model['name'] for model in list_models['models']]
|
|
except Exception as e1:
|
|
print(f"Error fetching models from 127.0.0.1: {e1}")
|
|
try:
|
|
# Fallback to 0.0.0.0
|
|
client = Client(host="http://0.0.0.0:11434")
|
|
list_models = client.list()
|
|
return [model['name'] for model in list_models['models']]
|
|
except Exception as e2:
|
|
print(f"Error fetching models from 0.0.0.0: {e2}")
|
|
return ["none"] # Return a default model if fetching fails
|
|
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
default_system_prompt = "Describe a specific example of an object, animal, person, or landscape based on a given general idea. Start with a clear and concise overall description in the first sentence. Then, provide a detailed depiction of its physical features, focusing on colors, size, clothing, eyes, and other distinguishing characteristics. Use commas to separate each detail and avoid listing them. Ensure each description is vivid, precise, and specific to one unique instance of the subject. Refrain from using poetic language and giving it a name.\nExample input: man\n Example output: \nAn overweight old man sitting on a bench, wearing a blue hat, yellow pants, orange jacket and black shirt, sunglasses, very long beard, very pale skin, long white hair, very large nose."
|
|
return {
|
|
"required": {
|
|
"user_prompt": ("STRING", {"multiline": True}),
|
|
"selected_model": (cls.get_available_models(),),
|
|
"system_prompt": ("STRING", {
|
|
"multiline": True,
|
|
"default": default_system_prompt
|
|
}),
|
|
"seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff}),
|
|
"keep_1min_in_vram": ("BOOLEAN", {"default": False})
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("STRING",)
|
|
RETURN_NAMES = ("ollama_response",)
|
|
FUNCTION = "connect_2_ollama"
|
|
CATEGORY = "Bjornulf"
|
|
|
|
def __init__(self):
|
|
self.last_content_hash = None
|
|
|
|
def connect_2_ollama(self, user_prompt, selected_model, system_prompt, keep_1min_in_vram, seed):
|
|
# Generate a hash of the current content
|
|
content_hash = hashlib.md5((user_prompt + selected_model + system_prompt).encode()).hexdigest()
|
|
|
|
# Check if the content has changed
|
|
if content_hash != self.last_content_hash:
|
|
# Content has changed, use the provided seed
|
|
self.last_content_hash = content_hash
|
|
else:
|
|
# Content hasn't changed, set seed to None to prevent randomization
|
|
seed = None
|
|
|
|
keep_alive_minutes = 0
|
|
if(keep_1min_in_vram):
|
|
keep_alive_minutes = 1
|
|
|
|
keep_alive = 0
|
|
# client = Client(host="http://0.0.0.0:11434")
|
|
# response = client.generate(
|
|
# model=selected_model,
|
|
# system=system_prompt,
|
|
# prompt=user_prompt,
|
|
# keep_alive=str(keep_alive_minutes) + "m"
|
|
# )
|
|
try:
|
|
# First attempt with 127.0.0.1
|
|
client = Client(host="http://127.0.0.1:11434")
|
|
response = client.generate(
|
|
model=selected_model,
|
|
system=system_prompt,
|
|
prompt=user_prompt,
|
|
keep_alive=str(keep_alive_minutes) + "m"
|
|
)
|
|
logging.info("Ollama response (127.0.0.1): " + response['response'])
|
|
except Exception as e:
|
|
logging.warning(f"Connection to 127.0.0.1 failed: {e}")
|
|
try:
|
|
# Fallback to 0.0.0.0 if 127.0.0.1 fails
|
|
client = Client(host="http://0.0.0.0:11434")
|
|
response = client.generate(
|
|
model=selected_model,
|
|
system=system_prompt,
|
|
prompt=user_prompt,
|
|
keep_alive=str(keep_alive_minutes) + "m"
|
|
)
|
|
logging.info("Ollama response (0.0.0.0): " + response['response'])
|
|
except Exception as e:
|
|
logging.error(f"Connection to 0.0.0.0 also failed: {e}")
|
|
logging.info("Ollama response : " + response['response'])
|
|
return (response['response'],) |