Spaces:
Runtime error
Runtime error
Update llm_local.py
Browse files- llm_local.py +7 -7
llm_local.py
CHANGED
|
@@ -63,13 +63,13 @@ def _mk(model):
|
|
| 63 |
|
| 64 |
|
| 65 |
WORKERS = {
|
| 66 |
-
"
|
| 67 |
"narrator": _mk(FAST_MODEL),
|
| 68 |
"reporter": _mk(FAST_MODEL),
|
| 69 |
"analyst": _mk(DEFAULT_MODEL),
|
| 70 |
}
|
| 71 |
# legacy aliases
|
| 72 |
-
_ALIAS = {"fast": "
|
| 73 |
|
| 74 |
|
| 75 |
def _wk(worker: str) -> str:
|
|
@@ -206,7 +206,7 @@ def load_model(name: str, worker: str = "analyst") -> str:
|
|
| 206 |
def auto_load_all():
|
| 207 |
"""Startup: tiny agents first (one small GGUF download serves all three),
|
| 208 |
then the Analyst. Runs in a background thread."""
|
| 209 |
-
for key in ("
|
| 210 |
load_model(WORKERS[key]["model"], worker=key)
|
| 211 |
|
| 212 |
|
|
@@ -219,7 +219,7 @@ def is_loaded(worker: str = None) -> bool:
|
|
| 219 |
|
| 220 |
def status() -> str:
|
| 221 |
lines = []
|
| 222 |
-
for key in ("
|
| 223 |
w = WORKERS[key]
|
| 224 |
label = WORKER_LABEL[key]
|
| 225 |
if w["llm"] is not None:
|
|
@@ -245,7 +245,7 @@ def _messages(user: str, system: str):
|
|
| 245 |
|
| 246 |
|
| 247 |
def chat(user: str, max_tokens: int = 500, temperature: float = 0.3,
|
| 248 |
-
system: str = DEFAULT_SYSTEM, worker: str = "
|
| 249 |
"""Blocking chat on one sub-agent (used by pipeline/agent code)."""
|
| 250 |
w = WORKERS[_wk(worker)]; worker = _wk(worker)
|
| 251 |
if w["llm"] is None:
|
|
@@ -265,7 +265,7 @@ def chat(user: str, max_tokens: int = 500, temperature: float = 0.3,
|
|
| 265 |
|
| 266 |
|
| 267 |
def chat_stream(user: str, max_tokens: int = 500, temperature: float = 0.3,
|
| 268 |
-
system: str = DEFAULT_SYSTEM, worker: str = "
|
| 269 |
"""Streaming chat on one sub-agent — yields cumulative text immediately."""
|
| 270 |
w = WORKERS[_wk(worker)]; worker = _wk(worker)
|
| 271 |
if w["llm"] is None:
|
|
@@ -297,7 +297,7 @@ def quick_test() -> str:
|
|
| 297 |
"""Sanity check both sub-agents."""
|
| 298 |
import time
|
| 299 |
outs = []
|
| 300 |
-
for key in ("
|
| 301 |
if WORKERS[key]["llm"] is None:
|
| 302 |
outs.append(f"{WORKER_LABEL[key]}: not loaded ({WORKERS[key]['stage']})")
|
| 303 |
continue
|
|
|
|
| 63 |
|
| 64 |
|
| 65 |
WORKERS = {
|
| 66 |
+
"interpreter": _mk(TRANSLATOR_MODEL),
|
| 67 |
"narrator": _mk(FAST_MODEL),
|
| 68 |
"reporter": _mk(FAST_MODEL),
|
| 69 |
"analyst": _mk(DEFAULT_MODEL),
|
| 70 |
}
|
| 71 |
# legacy aliases
|
| 72 |
+
_ALIAS = {"fast": "interpreter", "deep": "analyst"}
|
| 73 |
|
| 74 |
|
| 75 |
def _wk(worker: str) -> str:
|
|
|
|
| 206 |
def auto_load_all():
|
| 207 |
"""Startup: tiny agents first (one small GGUF download serves all three),
|
| 208 |
then the Analyst. Runs in a background thread."""
|
| 209 |
+
for key in ("interpreter", "narrator", "reporter", "analyst"):
|
| 210 |
load_model(WORKERS[key]["model"], worker=key)
|
| 211 |
|
| 212 |
|
|
|
|
| 219 |
|
| 220 |
def status() -> str:
|
| 221 |
lines = []
|
| 222 |
+
for key in ("interpreter", "narrator", "reporter", "analyst"):
|
| 223 |
w = WORKERS[key]
|
| 224 |
label = WORKER_LABEL[key]
|
| 225 |
if w["llm"] is not None:
|
|
|
|
| 245 |
|
| 246 |
|
| 247 |
def chat(user: str, max_tokens: int = 500, temperature: float = 0.3,
|
| 248 |
+
system: str = DEFAULT_SYSTEM, worker: str = "interpreter") -> str:
|
| 249 |
"""Blocking chat on one sub-agent (used by pipeline/agent code)."""
|
| 250 |
w = WORKERS[_wk(worker)]; worker = _wk(worker)
|
| 251 |
if w["llm"] is None:
|
|
|
|
| 265 |
|
| 266 |
|
| 267 |
def chat_stream(user: str, max_tokens: int = 500, temperature: float = 0.3,
|
| 268 |
+
system: str = DEFAULT_SYSTEM, worker: str = "interpreter"):
|
| 269 |
"""Streaming chat on one sub-agent — yields cumulative text immediately."""
|
| 270 |
w = WORKERS[_wk(worker)]; worker = _wk(worker)
|
| 271 |
if w["llm"] is None:
|
|
|
|
| 297 |
"""Sanity check both sub-agents."""
|
| 298 |
import time
|
| 299 |
outs = []
|
| 300 |
+
for key in ("interpreter", "narrator", "reporter", "analyst"):
|
| 301 |
if WORKERS[key]["llm"] is None:
|
| 302 |
outs.append(f"{WORKER_LABEL[key]}: not loaded ({WORKERS[key]['stage']})")
|
| 303 |
continue
|