serversdown
30185f3fd8
The MI50 box (CT202) runs an OpenAI-compatible llama.cpp server on 10.0.0.44:8080. Wire it in as a third backend: - llm.complete gains backend="mi50" (OpenAI client pointed at MI50_BASE_URL) - config: MI50_BASE_URL (default http://10.0.0.44:8080/v1) + MI50_MODEL - chat.respond labels the model per backend; web _backend_for maps "mi50" - UI backend selector adds "MI50 — local GPU" Verified end-to-end: llm.complete(backend="mi50") returns from the live server. See homelab-inference memory for the box topology. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
102 lines
3.9 KiB
Python
102 lines
3.9 KiB
Python
"""The chat turn loop: persona + tiered memory + recent context -> reply.
|
|
|
|
Context is assembled in tiers (oldest/most-compacted first):
|
|
1. persona
|
|
2. long-term gist — relevant *summaries* of other sessions
|
|
3. sharp details — a few raw cross-session exchanges (so specifics survive)
|
|
4. recent raw turns of the current session (full fidelity)
|
|
5. the new user message
|
|
After replying, the session is compacted if enough new turns have accumulated.
|
|
"""
|
|
from __future__ import annotations
|
|
|
|
from lyra import config, llm, logbus, memory, persona, summary
|
|
from lyra.llm import Backend, Message
|
|
|
|
RECALL_K = 3 # raw cross-session "sharp detail" hits
|
|
RECENT_N = 10 # raw turns of the current session
|
|
SUMMARY_K = 3 # other-session gists
|
|
|
|
|
|
def _summary_note(summaries: list[memory.Summary]) -> Message:
|
|
lines = [f"- ({s.created_at[:10]}) {s.content}" for s in summaries]
|
|
body = "Gist of earlier sessions (compacted — ask if you need specifics):\n" + "\n".join(lines)
|
|
return {"role": "system", "content": body}
|
|
|
|
|
|
def _detail_note(exchanges: list[memory.Exchange]) -> Message:
|
|
lines = [f"- ({ex.created_at[:10]}, {ex.role}) {ex.content}" for ex in exchanges]
|
|
body = "Specific things you recall from past conversations:\n" + "\n".join(lines)
|
|
return {"role": "system", "content": body}
|
|
|
|
|
|
def _render(messages: list[Message]) -> str:
|
|
"""Human-readable dump of the exact prompt, for the live-log inspector."""
|
|
return "\n\n".join(f"[{m['role']}]\n{m['content']}" for m in messages)
|
|
|
|
|
|
def build_messages(session_id: str, user_msg: str) -> list[Message]:
|
|
"""Assemble the full, tiered message list for one turn."""
|
|
messages: list[Message] = [{"role": "system", "content": persona.system_prompt()}]
|
|
|
|
# Semantic memory: the distilled profile (who Brian is) — answers identity
|
|
# questions that raw recall can't. Always in context when it exists.
|
|
profile = memory.get_profile()
|
|
if profile:
|
|
messages.append(
|
|
{"role": "system", "content": "What you know about Brian:\n" + profile}
|
|
)
|
|
|
|
recent = memory.recent(session_id, n=RECENT_N)
|
|
recent_ids = {ex.id for ex in recent}
|
|
|
|
# Tier 1: compacted gists of *other* sessions (long-term, general idea).
|
|
summaries = memory.recall_summaries(user_msg, k=SUMMARY_K, exclude_session=session_id)
|
|
if summaries:
|
|
messages.append(_summary_note(summaries))
|
|
|
|
# Tier 2: a few sharp raw details from other sessions (so specifics survive
|
|
# compaction). Skip the current session (its raw turns are in `recent`).
|
|
recalled = [
|
|
ex for ex in memory.recall(user_msg, k=RECALL_K)
|
|
if ex.id not in recent_ids and ex.session_id != session_id
|
|
]
|
|
if recalled:
|
|
messages.append(_detail_note(recalled))
|
|
|
|
# Tier 3: current session, full fidelity.
|
|
for ex in recent:
|
|
messages.append({"role": ex.role, "content": ex.content})
|
|
|
|
messages.append({"role": "user", "content": user_msg})
|
|
|
|
logbus.log(
|
|
"debug", "context built",
|
|
recent=len(recent), summaries=len(summaries), details=len(recalled),
|
|
chars=sum(len(m["content"]) for m in messages), detail=_render(messages),
|
|
)
|
|
return messages
|
|
|
|
|
|
def respond(session_id: str, user_msg: str, backend: Backend = "cloud") -> str:
|
|
"""Produce Lyra's reply to a single user message and persist the exchange."""
|
|
cfg = config.load()
|
|
model = {"local": cfg.local_model, "cloud": cfg.cloud_model, "mi50": cfg.mi50_model}.get(
|
|
backend, backend
|
|
)
|
|
logbus.log(
|
|
"info", "chat request", session=session_id, backend=backend,
|
|
model=model, embed=cfg.embed_backend,
|
|
)
|
|
|
|
messages = build_messages(session_id, user_msg)
|
|
reply = llm.complete(messages, backend=backend)
|
|
logbus.log("info", "reply", session=session_id, chars=len(reply))
|
|
|
|
memory.remember(session_id, "user", user_msg)
|
|
memory.remember(session_id, "assistant", reply)
|
|
|
|
# Compact this session once enough new turns have piled up.
|
|
summary.maybe_summarize(session_id)
|
|
return reply
|