Qwen3.8-Flash-Next on any consumer hardware: one-click install for Windows / Linux. Strata inference engine, OpenAI/Anthropic API on localhost, optional image input.
C++
"""The monitor's Overview (view 1): numbers from /metrics, the drawing, glyph fallbacks, and the monitor showing it."""
from __future__ import annotations
import asyncio
import io
import time
import unittest
from rich.console import Console
from strata_cli import overview
from strata_cli.monitor import MonitorApp
GIB = 2 ** 30
def metrics(state="reading", **over):
now = time.time()
hist = lambda base, amp: [base + amp * ((i * 7) % 5) / 4 for i in range(60)] # noqa: E731
m = {"time": now,
"engine": {"model": "IQ3_S", "max_context": 131072, "expert_slots": 1644, "experts": 5120, "expert_cache_mib": 3280},
"live": {"state": state, "queued": 0, "prompt_read": 35610 if state == "reading" else None, "prompt_total": 36742 if state == "reading" else None,
"prompt_tokens": 36742, "generated": 120 if state == "generating" else 0, "max_tokens": 4096,
"tok_s": 50.4 if state == "generating" else None, "prefill_tok_s_mean": 820.0},
"hardware": {"gpu_util": 88, "gpu_mem_used": 11.2 * GIB, "gpu_mem_total": 12 * GIB, "gpu_temp": 54, "gpu_power": 114, "gpu_power_limit": 250,
"gpu_pcie_rx_mb": 18237, "gpu_pcie_gen": 4, "gpu_pcie_width": 16, "cpu": 14, "ram_used": 58.5 * GIB, "ram_total": 63 * GIB,
"disk_read_mb": 0.1, "disk_write_mb": 0.0},
"hardware_static": {"gpu_name": "RTX 5070", "gpu_count": 1, "cores": 6, "threads": 12},
"history": {"tok_s": hist(40, 10), "gpu_util": hist(80, 10), "gpu_mem_used": [11.2 * GIB] * 60, "gpu_temp": [54] * 60,
"gpu_power": hist(110, 10), "gpu_pcie_rx_mb": hist(10000, 9000), "cpu": hist(10, 8), "disk_read_mb": hist(0, 3)},
"requests": [{"id": "a", "time": now - 5, "duration_s": 429.8, "finish_reason": "stop", "prompt_tokens": 17183, "reused": 0,
"output_tokens": 19511, "decode_ms": 386000},
{"id": "b", "time": now - 9, "duration_s": 4.0, "finish_reason": "stop", "prompt_tokens": 626, "reused": 0,
"output_tokens": 11, "decode_ms": 410}]}
m.update(over)
return m
def draw(model, width=110, height=30, **kw) -> str:
c = Console(width=width + 2, record=True, file=io.StringIO(), force_terminal=False, legacy_windows=False)
c.print(overview.render(model, width=width, height=height, connection="http://127.0.0.1:8080", **kw))
return c.export_text()
class Build(unittest.TestCase):
def test_reading_state_and_progress(self):
m = overview.build(metrics("reading"))
self.assertEqual(m["state"], "Reading")
self.assertEqual(m["progress"], (35610.0, 36742.0))
self.assertEqual(m["context"], (36742.0, 131072.0))
self.assertEqual(m["speed"][:2], (820.0, "reading the prompt"))
def test_generating_uses_the_live_rate_and_prompt_plus_generated_for_context(self):
m = overview.build(metrics("generating"))
self.assertEqual(m["state"], "Generating")
self.assertEqual(m["speed"][:2], (50.4, "now"))
self.assertEqual(m["context"][0], 36742 + 120)
self.assertEqual(m["progress"], (120.0, 4096.0))
def test_idle_shows_the_last_request(self):
m = overview.build(metrics("idle"))
self.assertEqual(m["state"], "Idle")
self.assertEqual(m["speed"][1], "last request")
self.assertAlmostEqual(m["speed"][0], 50.5, places=1) # 19511 tokens in 386 s
self.assertEqual(m["context"][0], 17183 + 19511)
self.assertEqual(m["progress"], (None, None))
def test_queued_and_error(self):
self.assertEqual(overview.build(metrics("idle", live={"state": "idle", "queued": 3}))["state"], "Queued")
self.assertEqual(overview.build(metrics("idle", engine={"model": "x", "error": "boom"}))["state"], "Error")
def test_hardware_in_gb(self):
m = overview.build(metrics())
self.assertAlmostEqual(m["gpu"]["vram_used"], 11.2)
self.assertEqual(m["gpu"]["vram_total"], 12.0)
self.assertAlmostEqual(m["ram"][0], 58.5)
self.assertEqual(m["experts"], {"slots": 1644.0, "total": 5120.0, "gb": 3280 / 1024.0})
def test_missing_numbers_stay_missing(self):
m = overview.build({"live": {"state": "idle"}, "engine": {}, "hardware": {}, "requests": []})
self.assertIsNone(m["gpu"]["util"])
self.assertIsNone(m["speed"][0])
text = draw(m)
self.assertIn("-- tok/s", text)
self.assertIn("No requests yet", text)
self.assertNotIn("0 %", text.replace("-- %", "")) # no invented zero
def test_a_server_without_metrics_gives_none(self):
self.assertIsNone(overview.build(None))
self.assertIsNone(overview.build({"engine": {}}))
class Draw(unittest.TestCase):
def test_the_approved_layout(self):
text = draw(overview.build(metrics("reading")), connect=True)
for part in ("STRATA Overview", "IQ3_S 128K", "Reading prompt · 97%", "[LIVE]", "[1 Overview]", "7 Connect",
"Model state", "[Reading]", "35,610 / 36,742 tokens · 97%",
"Speed", "GPU load", "VRAM", "GPU temp", "Power", "PCIe", "CPU", "Disk read",
"88 %", "RTX 5070", "11.2 / 12 GB", "1,644 experts cached", "54 °C", "114 W", "of 250 W limit", "Gen4 x16",
"to GPU 18,237 MB/s", "6 cores · 12 threads", "0.1 MB/s", "write 0.0 MB/s",
"Context fill", "36K / 128K", "Experts in VRAM", "1,644 · 3.2 GB", "System RAM", "58.5 / 63 GB",
"Recent requests", "TIME", "STATUS", "PROMPT", "REUSED", "OUTPUT", "TOK/S", "DURATION", "17,183", "19,511", "429.8 s",
"1-7 views", "b open in browser", "q quit"):
self.assertIn(part, text, part)
def test_without_connect_the_hint_says_six_views(self):
text = draw(overview.build(metrics()), connect=False)
self.assertIn("1-6 views", text)
self.assertNotIn("7 Connect", text)
def test_ascii_mode_has_no_glyphs_the_classic_windows_console_font_lacks(self):
text = draw(overview.build(metrics()), glyphs="ascii", color=False)
for bad in "▁▂▃▄▅▆▇█░╭╮╰╯✓●":
self.assertNotIn(bad, text, bad)
self.assertIn("#", text)
self.assertIn("54 C", text)
def test_fits_the_width_it_is_given(self):
for w in (80, 100, 120):
text = draw(overview.build(metrics()), width=w - 2)
self.assertTrue(all(len(line) <= w for line in text.splitlines()), w)
def test_spark(self):
self.assertEqual(len(overview.spark([1, 2, 3], 8, "unicode")), 8)
self.assertEqual(overview.spark([None, None], 4, "unicode"), " ")
self.assertEqual(overview.spark([0, 100], 2, "unicode", lo=0, hi=100), "▁█")
self.assertEqual(overview.spark([0, 100], 2, "ascii", lo=0, hi=100), ".@")
class Source:
root = "http://127.0.0.1:8095"
def __init__(self, raw):
self.raw = raw
self.base_url = self.root
def get(self, path):
if path == "/metrics":
return self.raw
raise AssertionError(path)
def snapshot(self):
return {"source": "fixture", "sampled_at": time.time(), "schema_version": 1, "instance": {"id": "x"},
"model": {"id": "IQ3_S", "loaded": True}, "work": {}, "rates": {}, "devices": [], "recent": [], "totals": {},
"topology": {}, "quality": {}}
class InTheMonitor(unittest.TestCase):
def test_view_one_is_the_new_overview_and_the_others_stay(self):
async def go():
app = MonitorApp(Source(metrics("generating")), None, interval=30)
async with app.run_test(size=(120, 36)) as pilot:
await pilot.pause(0.5)
shown = "\n".join(s.text for s in app.screen._compositor.render_strips())
self.assertIn("Model state", shown)
self.assertIn("Generating", shown)
await pilot.press("2")
await pilot.pause(0.2)
shown = "\n".join(s.text for s in app.screen._compositor.render_strips())
self.assertNotIn("Model state", shown)
self.assertIn("Serving", shown)
asyncio.run(go())
def test_an_older_server_without_metrics_keeps_the_earlier_overview(self):
class Old(Source):
def get(self, path):
from strata_cli.client import ClientError
raise ClientError("not found", http_status=404)
async def go():
app = MonitorApp(Old(None), None, interval=30)
async with app.run_test(size=(120, 36)) as pilot:
await pilot.pause(0.5)
shown = "\n".join(s.text for s in app.screen._compositor.render_strips())
self.assertNotIn("Model state", shown)
self.assertIn("MONITOR", shown)
asyncio.run(go())
if __name__ == "__main__":
unittest.main()