administrator/frigate
1
0
Fork 0
mirror of https://github.com/blakeblackshear/frigate.git synced 2026-07-15 00:11:15 +03:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

Pull correctly

Browse Source
This commit is contained in:
Nicolas Mowen 2026-04-02 17:37:47 -06:00
parent b7a4eb57e9
commit 0729af7246
1 changed files with 24 additions and 38 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

62
frigate/genai/llama_cpp.py
View File

@ -62,7 +62,7 @@ class LlamaCppClient(GenAIClient):
if base_url is None: if base_url is None:
return None return None
# Query /v1/models to validate the configured model exists # Query /v1/models to validate the model and extract capabilities
try: try:
response = requests.get( response = requests.get(
f"{base_url}/v1/models", f"{base_url}/v1/models",
@ -71,22 +71,17 @@ class LlamaCppClient(GenAIClient):
response.raise_for_status() response.raise_for_status()
models_data = response.json() models_data = response.json()
model_found = False matched_model = None
configured_model = self.genai_config.model configured_model = self.genai_config.model
for model in models_data.get("data", []): for model in models_data.get("data", []):
model_ids = {model.get("id")} model_ids = {model.get("id")}
# llama.cpp models can have aliases (e.g. short names)
for alias in model.get("aliases", []): for alias in model.get("aliases", []):
model_ids.add(alias) model_ids.add(alias)
if configured_model in model_ids: if configured_model in model_ids:
model_found = True matched_model = model
# Extract context size from model metadata if available
meta = model.get("meta", {})
if meta.get("n_ctx_train"):
self._context_size = int(meta["n_ctx_train"])
break break
if not model_found: if matched_model is None:
available = [] available = []
for m in models_data.get("data", []): for m in models_data.get("data", []):
available.append(m.get("id", "unknown")) available.append(m.get("id", "unknown"))
@ -98,36 +93,14 @@ class LlamaCppClient(GenAIClient):
available, available,
) )
return None return None
except Exception as e:
logger.warning(
"Failed to query llama.cpp /v1/models endpoint: %s. "
"Model validation skipped.",
e,
)
# Query /props for server properties (context size, modalities) # Parse capabilities from the model's server launch args
try: args = matched_model.get("status", {}).get("args", [])
response = requests.get( self._parse_model_args(args)
f"{base_url}/props",
timeout=10,
)
response.raise_for_status()
props = response.json()
# Context size from server runtime config
default_settings = props.get("default_generation_settings", {})
if default_settings.get("n_ctx"):
self._context_size = int(default_settings["n_ctx"])
# Check supported modalities
modalities = props.get("modalities", {})
self._supports_vision = modalities.get("vision", False)
self._supports_audio = modalities.get("audio", False)
self._supports_tools = modalities.get("tools", False)
logger.info( logger.info(
"llama.cpp model '%s' loaded — context: %s, vision: %s, audio: %s, tools: %s", "llama.cpp model '%s' initialized — context: %s, vision: %s, audio: %s, tools: %s",
self.genai_config.model, configured_model,
self._context_size or "unknown", self._context_size or "unknown",
self._supports_vision, self._supports_vision,
self._supports_audio, self._supports_audio,
@ -135,13 +108,26 @@ class LlamaCppClient(GenAIClient):
) )
except Exception as e: except Exception as e:
logger.warning( logger.warning(
"Failed to query llama.cpp /props endpoint: %s. " "Failed to query llama.cpp /v1/models endpoint: %s. "
"Using defaults for context size and modalities.", "Model validation skipped.",
e, e,
) )
return base_url return base_url
def _parse_model_args(self, args: list[str]) -> None:
"""Extract context size and capabilities from llama-server launch args."""
for i, arg in enumerate(args):
if arg == "--ctx-size" and i + 1 < len(args):
try:
self._context_size = int(args[i + 1])
except ValueError:
pass
elif arg == "--mmproj":
self._supports_vision = True
elif arg == "--jinja":
self._supports_tools = True
def _send( def _send(
self, self,
prompt: str, prompt: str,