mirror of
https://github.com/unslothai/unsloth.git
synced 2026-07-10 00:08:58 +00:00
9a966adf51
* Studio: trim serving-log noise and surface llama-server engine stats Studio prints one structured line per HTTP request, so the SPA's polling and per-invalidation fan-out bury the lines that matter. - Dedup identical successful GETs within a short window (default 300ms, UNSLOTH_STUDIO_ACCESS_LOG_DEDUP_MS) so a burst logs once. The dedup key includes the query string, so distinct query-driven GETs are not collapsed. Runs after the response is sent, so it adds no request latency; mutations, non-2xx, and loading polls are untouched. - Collapse pure-liveness polls (/api/health, /api/auth/status, /api/inference/status, /api/inference/monitor) to a longer heartbeat (default 10s, UNSLOTH_STUDIO_ACCESS_LOG_POLL_DEDUP_MS). The API monitor console polls /monitor every 1.5s while open. - Translate llama-server's Prometheus /metrics into a periodic vLLM-style engine_stats line (generation/prompt throughput and requests in flight) from a daemon poller, gated on UNSLOTH_STUDIO_ENGINE_STATS. Throughput uses llama-server's predicted_tokens_seconds / prompt_tokens_seconds gauges, with a tokens_predicted_total / prompt_tokens_total counter-delta fallback; it does not use n_decode_total (which counts llama_decode() calls, not tokens). No KV field is emitted, since llama.cpp does not expose kv_cache_usage_ratio. --metrics is added only when probe_server_capabilities reports the binary supports it, so older/custom binaries still load. The poller keeps retrying through transient scrape failures (stop() drives shutdown) and a malformed sample cannot crash its thread. - api_monitor.append_reply: once the preview cap is reached, skip the per-chunk re-concat (avoids O(n^2) on long generations) while still recording the "..." truncation marker for a reply that lands exactly on the cap. - unsloth studio --verbose and unsloth studio run --verbose both restore every per-request log; --verbose before a subcommand is rejected with guidance (matching --secure / --parallel). run --verbose still forwards --log-verbose to llama-server, preserving the pre-existing pass-through verbosity. * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
260 lines
7.8 KiB
Python
260 lines
7.8 KiB
Python
# SPDX-License-Identifier: AGPL-3.0-only
|
|
# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
|
|
|
|
from core.inference.api_monitor import ApiMonitor, _trim
|
|
|
|
|
|
def test_api_monitor_tracks_reply_usage_and_context():
|
|
monitor = ApiMonitor(max_entries = 3)
|
|
|
|
entry_id = monitor.start(
|
|
endpoint = "/v1/chat/completions",
|
|
method = "POST",
|
|
model = "local-model",
|
|
prompt = "user: hello",
|
|
context_length = 100,
|
|
)
|
|
monitor.append_reply(entry_id, "hi")
|
|
monitor.append_reply(entry_id, " there")
|
|
monitor.set_usage(
|
|
entry_id,
|
|
prompt_tokens = 4,
|
|
completion_tokens = 6,
|
|
)
|
|
monitor.finish(entry_id)
|
|
|
|
[entry] = monitor.snapshot()
|
|
assert entry["status"] == "completed"
|
|
assert entry["reply"] == "hi there"
|
|
assert entry["total_tokens"] == 10
|
|
assert entry["context_usage"] == 0.1
|
|
assert entry["duration_ms"] is not None
|
|
|
|
|
|
def test_api_monitor_summary_omits_full_prompt_and_reply():
|
|
monitor = ApiMonitor(max_entries = 3)
|
|
entry_id = monitor.start(
|
|
endpoint = "/v1/chat/completions",
|
|
method = "POST",
|
|
model = "local-model",
|
|
prompt = "p" * 500,
|
|
)
|
|
monitor.set_reply(entry_id, "r" * 500)
|
|
|
|
[summary] = monitor.snapshot(include_details = False)
|
|
assert "prompt" not in summary
|
|
assert "reply" not in summary
|
|
assert summary["prompt_preview"].endswith("...")
|
|
assert summary["reply_preview"].endswith("...")
|
|
assert summary["prompt_truncated"] is True
|
|
assert summary["reply_truncated"] is True
|
|
|
|
detail = monitor.get(entry_id)
|
|
assert detail is not None
|
|
assert detail["prompt"] == "p" * 500
|
|
assert detail["reply"] == "r" * 500
|
|
|
|
|
|
def test_api_monitor_filters_entries_by_subject():
|
|
monitor = ApiMonitor(max_entries = 3)
|
|
alice = monitor.start(
|
|
endpoint = "/v1/chat/completions",
|
|
method = "POST",
|
|
model = "m",
|
|
prompt = "alice prompt",
|
|
subject = "alice",
|
|
)
|
|
bob = monitor.start(
|
|
endpoint = "/v1/chat/completions",
|
|
method = "POST",
|
|
model = "m",
|
|
prompt = "bob prompt",
|
|
subject = "bob",
|
|
)
|
|
monitor.finish(bob)
|
|
|
|
alice_entries = monitor.snapshot(subject = "alice")
|
|
assert [entry["id"] for entry in alice_entries] == [alice]
|
|
assert monitor.get(bob, subject = "alice") is None
|
|
assert monitor.get(bob, subject = "bob")["id"] == bob
|
|
assert monitor.active_count(subject = "alice") == 1
|
|
assert monitor.active_count(subject = "bob") == 0
|
|
|
|
|
|
def test_api_monitor_keeps_bounded_recent_history():
|
|
monitor = ApiMonitor(max_entries = 2)
|
|
|
|
first = monitor.start(
|
|
endpoint = "/v1/chat/completions",
|
|
method = "POST",
|
|
model = "m",
|
|
prompt = "first",
|
|
)
|
|
second = monitor.start(
|
|
endpoint = "/v1/chat/completions",
|
|
method = "POST",
|
|
model = "m",
|
|
prompt = "second",
|
|
)
|
|
third = monitor.start(
|
|
endpoint = "/v1/chat/completions",
|
|
method = "POST",
|
|
model = "m",
|
|
prompt = "third",
|
|
)
|
|
monitor.finish(first)
|
|
monitor.finish(second)
|
|
monitor.finish(third)
|
|
|
|
entries = monitor.snapshot()
|
|
ids = [entry["id"] for entry in entries]
|
|
assert ids[0] == third
|
|
assert [entry["prompt"] for entry in entries] == ["third", "second"]
|
|
assert first not in ids
|
|
assert monitor.active_count() == 0
|
|
|
|
|
|
def test_api_monitor_keeps_running_entries_beyond_history_limit():
|
|
monitor = ApiMonitor(max_entries = 1)
|
|
|
|
running = monitor.start(
|
|
endpoint = "/v1/chat/completions",
|
|
method = "POST",
|
|
model = "m",
|
|
prompt = "running",
|
|
)
|
|
for prompt in ("done-1", "done-2", "done-3"):
|
|
entry_id = monitor.start(
|
|
endpoint = "/v1/chat/completions",
|
|
method = "POST",
|
|
model = "m",
|
|
prompt = prompt,
|
|
)
|
|
monitor.finish(entry_id)
|
|
|
|
entries = monitor.snapshot()
|
|
ids = [entry["id"] for entry in entries]
|
|
assert running in ids
|
|
assert monitor.active_count() == 1
|
|
|
|
monitor.finish(running)
|
|
[entry] = monitor.snapshot()
|
|
assert entry["id"] == running
|
|
assert entry["status"] == "completed"
|
|
assert monitor.active_count() == 0
|
|
|
|
|
|
def test_api_monitor_finish_is_idempotent():
|
|
monitor = ApiMonitor(max_entries = 2)
|
|
entry_id = monitor.start(
|
|
endpoint = "/v1/chat/completions",
|
|
method = "POST",
|
|
model = "m",
|
|
prompt = "hi",
|
|
)
|
|
monitor.finish(entry_id)
|
|
first = monitor.snapshot()[0]
|
|
monitor.finish(entry_id)
|
|
second = monitor.snapshot()[0]
|
|
assert first["finished_at"] == second["finished_at"]
|
|
assert first["duration_ms"] == second["duration_ms"]
|
|
|
|
|
|
def test_api_monitor_preserves_authoritative_total_tokens():
|
|
monitor = ApiMonitor(max_entries = 2)
|
|
entry_id = monitor.start(
|
|
endpoint = "/v1/chat/completions",
|
|
method = "POST",
|
|
model = "m",
|
|
prompt = "hi",
|
|
)
|
|
monitor.set_usage(
|
|
entry_id,
|
|
prompt_tokens = 10,
|
|
completion_tokens = 20,
|
|
total_tokens = 33,
|
|
)
|
|
# A later partial chunk omitting `total_tokens` must not clobber 33.
|
|
monitor.set_usage(entry_id, prompt_tokens = 11)
|
|
assert monitor.snapshot()[0]["total_tokens"] == 33
|
|
|
|
|
|
def test_api_monitor_recomputes_derived_total_tokens():
|
|
monitor = ApiMonitor(max_entries = 2)
|
|
entry_id = monitor.start(
|
|
endpoint = "/v1/chat/completions",
|
|
method = "POST",
|
|
model = "m",
|
|
prompt = "hi",
|
|
)
|
|
monitor.set_usage(entry_id, prompt_tokens = 10)
|
|
assert monitor.snapshot()[0]["total_tokens"] == 10
|
|
|
|
monitor.set_usage(entry_id, completion_tokens = 20)
|
|
entry = monitor.snapshot()[0]
|
|
assert entry["prompt_tokens"] == 10
|
|
assert entry["completion_tokens"] == 20
|
|
assert entry["total_tokens"] == 30
|
|
|
|
|
|
def test_api_monitor_duration_non_negative_under_clock_step(monkeypatch):
|
|
import core.inference.api_monitor as m
|
|
|
|
fake_now = [1000.0]
|
|
monkeypatch.setattr(m.time, "time", lambda: fake_now[0])
|
|
monitor = ApiMonitor(max_entries = 1)
|
|
entry_id = monitor.start(
|
|
endpoint = "/x",
|
|
method = "POST",
|
|
model = "m",
|
|
prompt = "hi",
|
|
)
|
|
fake_now[0] = 500.0
|
|
monitor.finish(entry_id)
|
|
assert monitor.snapshot()[0]["duration_ms"] >= 0
|
|
|
|
|
|
def test_api_monitor_trim_guards_tiny_limit():
|
|
assert _trim("abcdefgh", 2) == ".."
|
|
assert _trim("abcdefgh", 0) == ""
|
|
assert _trim("abcdefgh", 3) == "..."
|
|
assert _trim("abcdefgh", 4) == "a..."
|
|
assert _trim("abcdefgh", 100) == "abcdefgh"
|
|
|
|
|
|
def test_api_monitor_append_reply_caps_without_regrowing():
|
|
import core.inference.api_monitor as m
|
|
|
|
monitor = ApiMonitor(max_entries = 1)
|
|
entry_id = monitor.start(
|
|
endpoint = "/v1/chat/completions",
|
|
method = "POST",
|
|
model = "m",
|
|
prompt = "go",
|
|
)
|
|
monitor.append_reply(entry_id, "x" * (m._MAX_REPLY_CHARS + 500))
|
|
capped = monitor.snapshot()[0]["reply"]
|
|
assert len(capped) == m._MAX_REPLY_CHARS and capped.endswith("...")
|
|
|
|
# Chunks past the cap must not change or grow the stored preview.
|
|
monitor.append_reply(entry_id, "y" * 1000)
|
|
assert monitor.snapshot()[0]["reply"] == capped
|
|
|
|
|
|
def test_api_monitor_append_reply_exact_cap_then_more_marks_truncated():
|
|
import core.inference.api_monitor as m
|
|
|
|
monitor = ApiMonitor(max_entries = 1)
|
|
entry_id = monitor.start(
|
|
endpoint = "/v1/chat/completions",
|
|
method = "POST",
|
|
model = "m",
|
|
prompt = "go",
|
|
)
|
|
# A reply landing exactly on the cap has no "..." marker yet.
|
|
monitor.append_reply(entry_id, "x" * m._MAX_REPLY_CHARS)
|
|
assert not monitor.snapshot()[0]["reply"].endswith("...")
|
|
# One more chunk must record the truncation, not silently freeze.
|
|
monitor.append_reply(entry_id, "y")
|
|
reply = monitor.snapshot()[0]["reply"]
|
|
assert len(reply) == m._MAX_REPLY_CHARS and reply.endswith("...")
|