onwx' (onwx-dash) 1.2.3: live speeds while an answer runs, animated 2-PC link, model picker without the path (part 2)
Browse files- onwx/chat.py +21 -0
- onwx/i18n.json +6 -2
- onwx/manager.py +18 -1
- onwx/server.py +11 -0
- onwx/web/manage.html +65 -14
onwx/chat.py
CHANGED
|
@@ -541,6 +541,23 @@ class ChatEngine:
|
|
| 541 |
x = y
|
| 542 |
|
| 543 |
# ---- API ----
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 544 |
def stream_chat(self, messages, max_new_tokens=512, temperature=0.0, top_p=1.0, stop=(), seed=None,
|
| 545 |
think=None, should_stop=None, top_k=0, repetition_penalty=1.0, presence_penalty=0.0,
|
| 546 |
frequency_penalty=0.0, tools=None, split_reasoning=False, show_thinking=None):
|
|
@@ -607,6 +624,9 @@ class ChatEngine:
|
|
| 607 |
# the client may leave during a long prompt (minutes on a big one, every other request waiting): stopped between
|
| 608 |
# blocks. The checkpoint stays right - its rows are below where this prompt writes (a fresh prompt dropped it)
|
| 609 |
self._cancel = should_stop
|
|
|
|
|
|
|
|
|
|
| 610 |
self.model.feeding = True # ([MTP] the MTP layer's rows go along with the prompt)
|
| 611 |
w = int(os.environ.get("ONWX_MTP_ROWS", "128") or 0) # ([MTP] rows of the last w prompt tokens only (~3.5 ms a row); 0: all)
|
| 612 |
self.model.mtp_from = max(0, n - w) if w > 0 else 0
|
|
@@ -679,6 +699,7 @@ class ChatEngine:
|
|
| 679 |
stop = [x for x in stop if x]
|
| 680 |
limit = min(max_new_tokens, self.max_context - n - 16)
|
| 681 |
t1 = time.time()
|
|
|
|
| 682 |
|
| 683 |
def deltas(final=False):
|
| 684 |
"""Send what is new. Held back until the next token (or the end): a tail that may still become a stop
|
|
|
|
| 541 |
x = y
|
| 542 |
|
| 543 |
# ---- API ----
|
| 544 |
+
live = None # [1.2.3] the request in progress (see live_now)
|
| 545 |
+
|
| 546 |
+
def live_now(self):
|
| 547 |
+
"""[1.2.3] the request in progress for the window, read from another thread: {"phase": "prompt", "new"
|
| 548 |
+
(tokens to read), "done" (read so far, where the model's state grows by chunk), "tok_s" (0 until a chunk is
|
| 549 |
+
done), "s"} or {"phase": "gen", "tokens", "tok_s", "s", "pre" (that prompt's tok/s)}, or None."""
|
| 550 |
+
lv, now = self.live, time.time()
|
| 551 |
+
if not lv:
|
| 552 |
+
return None
|
| 553 |
+
if lv["phase"] == "prompt":
|
| 554 |
+
s = now - lv["t0"]
|
| 555 |
+
done = max(0, min(lv["new"], len(getattr(self.model, "held", None) or []) - lv["h0"]))
|
| 556 |
+
return {"phase": "prompt", "new": lv["new"], "done": done, "tok_s": done / s if done and s > 0 else 0,
|
| 557 |
+
"s": s}
|
| 558 |
+
s, k = now - lv["t1"], len(lv["out"])
|
| 559 |
+
return {"phase": "gen", "tokens": k, "tok_s": k / s if k and s > 0.2 else 0, "s": s, "pre": lv["pre"]}
|
| 560 |
+
|
| 561 |
def stream_chat(self, messages, max_new_tokens=512, temperature=0.0, top_p=1.0, stop=(), seed=None,
|
| 562 |
think=None, should_stop=None, top_k=0, repetition_penalty=1.0, presence_penalty=0.0,
|
| 563 |
frequency_penalty=0.0, tools=None, split_reasoning=False, show_thinking=None):
|
|
|
|
| 624 |
# the client may leave during a long prompt (minutes on a big one, every other request waiting): stopped between
|
| 625 |
# blocks. The checkpoint stays right - its rows are below where this prompt writes (a fresh prompt dropped it)
|
| 626 |
self._cancel = should_stop
|
| 627 |
+
# [1.2.3] what the window shows live (server /api/live): reading the prompt, then generating
|
| 628 |
+
h0 = len(getattr(self.model, "held", None) or [])
|
| 629 |
+
self.live = {"phase": "prompt", "t0": time.time(), "new": n - reuse, "h0": h0}
|
| 630 |
self.model.feeding = True # ([MTP] the MTP layer's rows go along with the prompt)
|
| 631 |
w = int(os.environ.get("ONWX_MTP_ROWS", "128") or 0) # ([MTP] rows of the last w prompt tokens only (~3.5 ms a row); 0: all)
|
| 632 |
self.model.mtp_from = max(0, n - w) if w > 0 else 0
|
|
|
|
| 699 |
stop = [x for x in stop if x]
|
| 700 |
limit = min(max_new_tokens, self.max_context - n - 16)
|
| 701 |
t1 = time.time()
|
| 702 |
+
self.live = {"phase": "gen", "t1": t1, "out": out, "pre": (n - reuse) / max(t_pre, 1e-9) if n - reuse >= 16 else 0}
|
| 703 |
|
| 704 |
def deltas(final=False):
|
| 705 |
"""Send what is new. Held back until the next token (or the end): a tail that may still become a stop
|
onwx/i18n.json
CHANGED
|
@@ -545,7 +545,9 @@
|
|
| 545 |
"メモリに読み込み中": "Loading into memory",
|
| 546 |
"{0} / 約 {1} GB": "{0} / about {1} GB",
|
| 547 |
"メイン PC からモデルの半分を受け取って、メモリに読み込んでいます": "Receiving half of the model from the main PC and loading it into memory",
|
| 548 |
-
"自分のモデルファイルからモデルの半分を読み込んでいます": "Loading half of the model from its own model file"
|
|
|
|
|
|
|
| 549 |
},
|
| 550 |
"zh": {
|
| 551 |
"2 台の PC を USB4(Thunderbolt)のケーブルでつなぎ、1 つのモデルを半分ずつ持って動かします。モデルのメモリが半分になり、1 台では載らないモデルも動き、生成も速くなります(Ryzen AI 9 HX のミニ PC 2 台で、Nemotron 3.5 Bolt 74B-A9B は生成が 1 台の約 1.55 倍・プロンプトの読み込みが約 1.4〜1.5 倍、Qwen3.8 27B は生成が約 1.7 倍、Qwen3.6 35B-A3B は約 1.4 倍)。モデルはメイン PC にあれば動きます。使えるのは Nemotron 3.5 Bolt、Qwen3.6 35B-A3B、Qwen3.8 27B の GGUF で、GPU だけで動かします(NPU の手伝いと画像はこのモードでは使えません)。2 台とも Linux で使ってください。": "用 USB4(Thunderbolt)线连接两台电脑,每台各持模型的一半来运行一个模型。每台所需内存减半,一台装不下的模型也能运行,生成也更快(两台 Ryzen AI 9 HX 迷你电脑:Nemotron 3.5 Bolt 74B-A9B 的生成约为一台的 1.55 倍、读取提示词约 1.4〜1.5 倍,Qwen3.8 27B 生成约 1.7 倍,Qwen3.6 35B-A3B 约 1.4 倍)。模型只需放在主电脑上。可用的是 Nemotron 3.5 Bolt、Qwen3.6 35B-A3B、Qwen3.8 27B 的 GGUF,只用 GPU 运行(此模式下不能使用 NPU 协助和图片)。两台都请使用 Linux。",
|
|
@@ -1046,6 +1048,8 @@
|
|
| 1046 |
"メモリに読み込み中": "正在载入内存",
|
| 1047 |
"{0} / 約 {1} GB": "{0} / 约 {1} GB",
|
| 1048 |
"メイン PC からモデルの半分を受け取って、メモリに読み込んでいます": "正在从主电脑接收一半模型并载入内存",
|
| 1049 |
-
"自分のモデルファイルからモデルの半分を読み込んでいます": "正在从自己的模型文件载入一半模型"
|
|
|
|
|
|
|
| 1050 |
}
|
| 1051 |
}
|
|
|
|
| 545 |
"メモリに読み込み中": "Loading into memory",
|
| 546 |
"{0} / 約 {1} GB": "{0} / about {1} GB",
|
| 547 |
"メイン PC からモデルの半分を受け取って、メモリに読み込んでいます": "Receiving half of the model from the main PC and loading it into memory",
|
| 548 |
+
"自分のモデルファイルからモデルの半分を読み込んでいます": "Loading half of the model from its own model file",
|
| 549 |
+
"● 生成中 {0}": "● generating {0}",
|
| 550 |
+
"● 読み込み中 {0} 秒": "● reading {0} s"
|
| 551 |
},
|
| 552 |
"zh": {
|
| 553 |
"2 台の PC を USB4(Thunderbolt)のケーブルでつなぎ、1 つのモデルを半分ずつ持って動かします。モデルのメモリが半分になり、1 台では載らないモデルも動き、生成も速くなります(Ryzen AI 9 HX のミニ PC 2 台で、Nemotron 3.5 Bolt 74B-A9B は生成が 1 台の約 1.55 倍・プロンプトの読み込みが約 1.4〜1.5 倍、Qwen3.8 27B は生成が約 1.7 倍、Qwen3.6 35B-A3B は約 1.4 倍)。モデルはメイン PC にあれば動きます。使えるのは Nemotron 3.5 Bolt、Qwen3.6 35B-A3B、Qwen3.8 27B の GGUF で、GPU だけで動かします(NPU の手伝いと画像はこのモードでは使えません)。2 台とも Linux で使ってください。": "用 USB4(Thunderbolt)线连接两台电脑,每台各持模型的一半来运行一个模型。每台所需内存减半,一台装不下的模型也能运行,生成也更快(两台 Ryzen AI 9 HX 迷你电脑:Nemotron 3.5 Bolt 74B-A9B 的生成约为一台的 1.55 倍、读取提示词约 1.4〜1.5 倍,Qwen3.8 27B 生成约 1.7 倍,Qwen3.6 35B-A3B 约 1.4 倍)。模型只需放在主电脑上。可用的是 Nemotron 3.5 Bolt、Qwen3.6 35B-A3B、Qwen3.8 27B 的 GGUF,只用 GPU 运行(此模式下不能使用 NPU 协助和图片)。两台都请使用 Linux。",
|
|
|
|
| 1048 |
"メモリに読み込み中": "正在载入内存",
|
| 1049 |
"{0} / 約 {1} GB": "{0} / 约 {1} GB",
|
| 1050 |
"メイン PC からモデルの半分を受け取って、メモリに読み込んでいます": "正在从主电脑接收一半模型并载入内存",
|
| 1051 |
+
"自分のモデルファイルからモデルの半分を読み込んでいます": "正在从自己的模型文件载入一半模型",
|
| 1052 |
+
"● 生成中 {0}": "● 生成中 {0}",
|
| 1053 |
+
"● 読み込み中 {0} 秒": "● 读取中 {0} 秒"
|
| 1054 |
}
|
| 1055 |
}
|
onwx/manager.py
CHANGED
|
@@ -251,6 +251,20 @@ class Manager:
|
|
| 251 |
return {"error": T("インストールの途中です")}
|
| 252 |
return {}
|
| 253 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 254 |
SPEED = [0, []] # (speed.json's time, its content)
|
| 255 |
|
| 256 |
@classmethod
|
|
@@ -716,7 +730,10 @@ class _Handler(BaseHTTPRequestHandler):
|
|
| 716 |
if u.path == "/api/state":
|
| 717 |
if not self._authed(q):
|
| 718 |
return self._send(403, {"error": T("この画面は onwx-dash のウィンドウかトレイから開いてください")})
|
| 719 |
-
|
|
|
|
|
|
|
|
|
|
| 720 |
self._send(404, {"error": "not found"})
|
| 721 |
|
| 722 |
def do_POST(self):
|
|
|
|
| 251 |
return {"error": T("インストールの途中です")}
|
| 252 |
return {}
|
| 253 |
|
| 254 |
+
@staticmethod
|
| 255 |
+
def live(st):
|
| 256 |
+
"""[1.2.3] the running server's request in progress (its /api/live), or None - asked only while the window's
|
| 257 |
+
server page shows (once a second)."""
|
| 258 |
+
sv = st.get("server") or {}
|
| 259 |
+
if sv.get("status") != "running" or not sv.get("port"):
|
| 260 |
+
return None
|
| 261 |
+
try:
|
| 262 |
+
import urllib.request
|
| 263 |
+
with urllib.request.urlopen(f"http://127.0.0.1:{sv['port']}/api/live", timeout=0.4) as r:
|
| 264 |
+
return json.load(r).get("live")
|
| 265 |
+
except Exception:
|
| 266 |
+
return None
|
| 267 |
+
|
| 268 |
SPEED = [0, []] # (speed.json's time, its content)
|
| 269 |
|
| 270 |
@classmethod
|
|
|
|
| 730 |
if u.path == "/api/state":
|
| 731 |
if not self._authed(q):
|
| 732 |
return self._send(403, {"error": T("この画面は onwx-dash のウィンドウかトレイから開いてください")})
|
| 733 |
+
st = self.manager.state()
|
| 734 |
+
if (q.get("live") or [""])[0] == "1": # [1.2.3] the server page shows: the answer in progress
|
| 735 |
+
st["live"] = self.manager.live(st)
|
| 736 |
+
return self._send(200, st)
|
| 737 |
self._send(404, {"error": "not found"})
|
| 738 |
|
| 739 |
def do_POST(self):
|
onwx/server.py
CHANGED
|
@@ -263,6 +263,7 @@ def start_generation(msgs, kw):
|
|
| 263 |
put(d)
|
| 264 |
finally:
|
| 265 |
set_priority(False)
|
|
|
|
| 266 |
except Exception as e: # surfaced to the client
|
| 267 |
put(e)
|
| 268 |
finally:
|
|
@@ -529,6 +530,16 @@ def bench_text(tokens: int = 2048):
|
|
| 529 |
return {"text": pipe.tok.decode(ids, skip_special_tokens=True), "tokens": len(ids)}
|
| 530 |
|
| 531 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 532 |
@app.get("/api/info")
|
| 533 |
def api_info():
|
| 534 |
# the window's defaults for this model, read now (they change without a restart)
|
|
|
|
| 263 |
put(d)
|
| 264 |
finally:
|
| 265 |
set_priority(False)
|
| 266 |
+
pipe.live = None # [1.2.3] (nothing in progress for the window)
|
| 267 |
except Exception as e: # surfaced to the client
|
| 268 |
put(e)
|
| 269 |
finally:
|
|
|
|
| 530 |
return {"text": pipe.tok.decode(ids, skip_special_tokens=True), "tokens": len(ids)}
|
| 531 |
|
| 532 |
|
| 533 |
+
@app.get("/api/live")
|
| 534 |
+
def api_live():
|
| 535 |
+
"""[1.2.3] the request in progress, for the window's speed tiles (polled once a second while its server page
|
| 536 |
+
shows): {"live": None | {"phase": "prompt" | "gen", ...}} (ChatEngine.live_now)."""
|
| 537 |
+
try:
|
| 538 |
+
return {"live": pipe.live_now() if pipe is not None and hasattr(pipe, "live_now") else None}
|
| 539 |
+
except Exception:
|
| 540 |
+
return {"live": None}
|
| 541 |
+
|
| 542 |
+
|
| 543 |
@app.get("/api/info")
|
| 544 |
def api_info():
|
| 545 |
# the window's defaults for this model, read now (they change without a restart)
|
onwx/web/manage.html
CHANGED
|
@@ -105,6 +105,32 @@
|
|
| 105 |
.wire.on::after { background: repeating-linear-gradient(90deg, #c44dff 0 8px, transparent 8px 12px); }
|
| 106 |
.wire.wait::after { background: repeating-linear-gradient(90deg, #f0c060 0 4px, transparent 4px 12px); }
|
| 107 |
.wire span { position: relative; z-index: 1; background: var(--solid); padding: 0 8px; font: 10px var(--mono); letter-spacing: .12em; color: var(--muted); }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 108 |
.kv { display: grid; grid-template-columns: max-content 1fr; gap: 3px 12px; align-items: center; }
|
| 109 |
.kv .k { color: var(--muted); }
|
| 110 |
.kv code { font-family: var(--mono); font-size: 12px; background: var(--panel2); color: #a9b8ff; padding: 1px 6px; border-radius: 2px; }
|
|
@@ -212,9 +238,9 @@
|
|
| 212 |
<section class="page on" id="page-server">
|
| 213 |
<!-- [1.2] the telemetry strip: the last answers' speeds, memory, the NPU's load (or the context length) -->
|
| 214 |
<div class="strip">
|
| 215 |
-
<div class="cell main" id="c-gen"><span class="lab"><i>01</i><span>生成</span></span><span class="num g off" id="t-gen">—</span><span class="unit">tok/s</span>
|
| 216 |
<svg class="spark" viewBox="0 0 100 22" preserveAspectRatio="none" aria-hidden="true"><defs><linearGradient id="sg" gradientUnits="userSpaceOnUse" x1="0" y1="0" x2="100" y2="0"><stop offset="0" stop-color="#ff4d5e"/><stop offset=".55" stop-color="#c44dff"/><stop offset="1" stop-color="#4d7dff"/></linearGradient></defs><polyline id="sp-gen" fill="none" stroke="url(#sg)" stroke-width="1.5" stroke-linejoin="round" vector-effect="non-scaling-stroke"/></svg></div>
|
| 217 |
-
<div class="cell" id="c-pre"><span class="lab"><i>02</i><span>プロンプト</span></span><span class="num off" id="t-pre">—</span><span class="unit">tok/s</span>
|
| 218 |
<svg class="spark" viewBox="0 0 100 22" preserveAspectRatio="none" aria-hidden="true"><polyline id="sp-pre" fill="none" stroke="#8fb0ff" stroke-width="1.5" stroke-linejoin="round" vector-effect="non-scaling-stroke"/></svg></div>
|
| 219 |
<div class="cell" id="c-mem"><span class="lab"><i>03</i><span>メモリ</span></span><span class="num off" id="t-mem">—</span><span class="unit" id="t-memu">GB</span>
|
| 220 |
<span class="gauge" title="onwx-dash(サーバーと、NPU が使うメモリ)・ほかのアプリ・空き"><i class="u-onwx" id="u-onwx"></i><i class="u-oth" id="u-oth"></i></span>
|
|
@@ -250,9 +276,9 @@
|
|
| 250 |
<button class="primary" id="b-tp" onclick="tpToggle()">補助を開始</button>
|
| 251 |
</div>
|
| 252 |
<div class="link">
|
| 253 |
-
<div class="pc me"><span class="lab" id="tp-me-role">—</span><b id="tp-me">—</b></div>
|
| 254 |
-
<div class="wire" id="tp-wire"><span id="tp-addr">—</span></div>
|
| 255 |
-
<div class="pc"><span class="lab" id="tp-peer-role">—</span><b id="tp-peer">—</b></div>
|
| 256 |
</div>
|
| 257 |
<div class="kv small" id="tp-modelrow">
|
| 258 |
<span class="k">モデル</span><span class="mono ellipsis" id="tp-model">—</span>
|
|
@@ -465,7 +491,7 @@
|
|
| 465 |
const cur = loadedNow && inst.some(m => m.key === sv.model) ? sv.model
|
| 466 |
: (pick.value && inst.some(m => m.key === pick.value)) ? pick.value
|
| 467 |
: (sv.model && inst.some(m => m.key === sv.model) ? sv.model : (inst.find(m => m.key === s.config.model) || inst[0] || {}).key);
|
| 468 |
-
const pickHtml = inst.map(m => `<option value="${esc(m.key)}"${m.key === cur ? " selected" : ""}
|
| 469 |
if (pickHtml !== lastPick) { pick.innerHTML = pickHtml; lastPick = pickHtml; } // rebuild only on change: a rebuild eats clicks
|
| 470 |
if (loadedNow && cur && pick.value !== cur) pick.value = cur;
|
| 471 |
const same = sv.model && pick.value === sv.model;
|
|
@@ -538,7 +564,7 @@
|
|
| 538 |
$("f-tp_enabled").checked = !!t.enabled; $("f-tp_role").value = t.role; $("f-tp_listener").value = t.listener;
|
| 539 |
$("f-tp_mode").value = t.ip === "auto" ? "auto" : tpRing(t.ip) ? "ring" : "manual"; $("f-tp_ip").value = t.ip === "auto" || tpRing(t.ip) ? "" : t.ip; $("f-tp_port").value = t.port; $("f-tp_streams").value = t.streams;
|
| 540 |
const opts = [`<option value="">${esc(T("自動(選んでいるモデルか、最初に見つかったもの)"))}</option>`]
|
| 541 |
-
.concat((s.tp.models || []).map(m => `<option value="${esc(m.key)}"
|
| 542 |
if (t.model && !(s.tp.models || []).some(m => m.key === t.model)) opts.push(`<option value="${esc(t.model)}">${esc(t.model)}</option>`);
|
| 543 |
const html = opts.join("");
|
| 544 |
if ($("f-tp_model").dataset.html !== html) { $("f-tp_model").innerHTML = html; $("f-tp_model").dataset.html = html; }
|
|
@@ -559,13 +585,21 @@
|
|
| 559 |
function setNum(id, text, on) { const e = $(id); if (e.textContent !== text) e.textContent = text; e.classList.toggle("off", !on); }
|
| 560 |
function renderStrip() {
|
| 561 |
const s = S, sv = s.server, live = sv.status === "running";
|
| 562 |
-
const sp = (live && s.speed) || [], key = JSON.stringify(sp) + live;
|
| 563 |
if (key !== lastSpeed) {
|
| 564 |
lastSpeed = key;
|
| 565 |
const gen = sp.map(x => x[0]).filter(x => x > 0), pre = sp.map(x => x[1]).filter(x => x > 0);
|
| 566 |
-
|
| 567 |
-
|
| 568 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 569 |
$("c-gen").title = $("c-pre").title = T("直近の回答の速さ。線は最近の回答の移り変わり");
|
| 570 |
}
|
| 571 |
// memory: onwx (the server and what the NPU holds for it) / other programs / free
|
|
@@ -676,7 +710,11 @@
|
|
| 676 |
: stg === "load" ? ["warn", T("メモリに読み込み中 {0}", pc(ld.gpu, ld.main_total))]
|
| 677 |
: stg === "helper" ? ["warn", T("接続しました")] : ["warn", T("補助 PC を待っています")];
|
| 678 |
$("tp-pill").className = "pill " + pl[0]; $("tp-pill").textContent = pl[1];
|
| 679 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 680 |
$("tp-me-role").textContent = T("この PC ・ メイン"); $("tp-peer-role").textContent = T("補助 PC");
|
| 681 |
$("tp-me").textContent = on || tw ? (sv.title || "—") : "—";
|
| 682 |
$("tp-addr").textContent = tpShow(t.ip, t.port); $("tp-wire").title = tpAddr(t.listener === "main", t.ip, t.port);
|
|
@@ -694,7 +732,10 @@
|
|
| 694 |
const pl = {waiting: ["warn", T("メイン PC を待っています")], serving: ["ok", T("メイン PC とつながっています")],
|
| 695 |
failed: ["danger", T("補助が止まりました")], stopped: ["", T("補助は停止中")]}[h.status] || ["", h.status];
|
| 696 |
$("tp-pill").className = "pill " + pl[0]; $("tp-pill").textContent = pl[1];
|
| 697 |
-
$("tp-wire").className = "wire" + (h.status === "serving" ? " on" : h.status === "waiting" ? " wait" : "")
|
|
|
|
|
|
|
|
|
|
| 698 |
$("tp-me-role").textContent = T("この PC ・ 補助"); $("tp-peer-role").textContent = T("メイン PC");
|
| 699 |
const hl = h.status === "serving" && h.loading; // [1.2.2] the helper's half still loading
|
| 700 |
if (hl) { $("tp-pill").className = "pill warn";
|
|
@@ -715,6 +756,14 @@
|
|
| 715 |
: T("「補助を開始」を押すと、この PC のモデルを閉じてメイン PC を待ちます");
|
| 716 |
$("tp-msg").textContent = msg; $("tp-msg").className = "small grow " + (h.error ? "err" : "muted");
|
| 717 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 718 |
async function tpToggle() {
|
| 719 |
const h = S && S.tp && S.tp.helper, up = h && (h.status === "waiting" || h.status === "serving");
|
| 720 |
if (!up && S.server.model && !await ask(T("補助を開始すると、読み込まれているモデル({0})を閉じます。よろしいですか?", S.server.title || S.server.model))) return;
|
|
@@ -939,7 +988,9 @@
|
|
| 939 |
async function refresh() {
|
| 940 |
const n = ++REQ;
|
| 941 |
try {
|
| 942 |
-
|
|
|
|
|
|
|
| 943 |
if (r.status === 403) { $("env").textContent = T("この画面は onwx-dash のウィンドウかトレイから開いてください"); return; }
|
| 944 |
const j = await r.json();
|
| 945 |
if (n <= STALE) return; // (asked before the last save: shows the old values)
|
|
|
|
| 105 |
.wire.on::after { background: repeating-linear-gradient(90deg, #c44dff 0 8px, transparent 8px 12px); }
|
| 106 |
.wire.wait::after { background: repeating-linear-gradient(90deg, #f0c060 0 4px, transparent 4px 12px); }
|
| 107 |
.wire span { position: relative; z-index: 1; background: var(--solid); padding: 0 8px; font: 10px var(--mono); letter-spacing: .12em; color: var(--muted); }
|
| 108 |
+
/* [1.2.3] the link alive: packets along the wire (both ways while the pair works, one way while the model is sent),
|
| 109 |
+
an amber scanner while waiting; the two PCs glow while joined, a bar while their half loads */
|
| 110 |
+
.wire { overflow: hidden; }
|
| 111 |
+
.wire.on::after { box-shadow: 0 0 6px #c44dff88; }
|
| 112 |
+
.wire .pk { position: absolute; top: 50%; left: -24px; width: 22px; height: 4px; margin-top: -2px; border-radius: 2px; opacity: 0;
|
| 113 |
+
background: linear-gradient(90deg, transparent, #c44dff 60%, #fff); box-shadow: 0 0 8px #c44dffaa; }
|
| 114 |
+
.wire .pk.b { background: linear-gradient(270deg, transparent, #4d7dff 60%, #fff); box-shadow: 0 0 8px #4d7dffaa; }
|
| 115 |
+
@keyframes pk-f { 0% { left: -24px; opacity: 0 } 12% { opacity: 1 } 88% { opacity: 1 } 100% { left: 100%; opacity: 0 } }
|
| 116 |
+
@keyframes pk-b { 0% { left: 100%; opacity: 0 } 12% { opacity: 1 } 88% { opacity: 1 } 100% { left: -24px; opacity: 0 } }
|
| 117 |
+
@keyframes pk-scan { 0% { left: 0; opacity: .25 } 50% { opacity: 1 } 100% { left: calc(100% - 22px); opacity: .25 } }
|
| 118 |
+
.wire.on .pk.f { animation: pk-f 2.8s linear var(--d) infinite; } .wire.on .pk.b { animation: pk-b 2.8s linear var(--d) infinite; }
|
| 119 |
+
.wire.on.busy .pk.f, .wire.on.busy .pk.b { animation-duration: .75s; }
|
| 120 |
+
.wire.on.send .pk.f { animation-duration: .7s; } .wire.on.send .pk.b { animation: none; }
|
| 121 |
+
.wire.on.recv .pk.b { animation-duration: .7s; } .wire.on.recv .pk.f { animation: none; }
|
| 122 |
+
.wire.wait .pk.f.s0 { animation: pk-scan 1.6s ease-in-out infinite alternate; background: radial-gradient(closest-side, #ffd27a, #f0c06000); box-shadow: none; height: 6px; margin-top: -3px; }
|
| 123 |
+
@keyframes pc-glow { 0%, 100% { box-shadow: 0 0 0 1px #c44dff22, 0 0 6px #c44dff22 } 50% { box-shadow: 0 0 0 1px #c44dff55, 0 0 16px #c44dff55 } }
|
| 124 |
+
.pc { position: relative; transition: border-color .4s; }
|
| 125 |
+
.pc.lit { border-color: #8a4dd0; animation: pc-glow 2.8s ease-in-out infinite; }
|
| 126 |
+
.pc.wait { border-style: dashed; }
|
| 127 |
+
.pc .bar { display: none; height: 3px; margin-top: 4px; background: var(--panel2); overflow: hidden; }
|
| 128 |
+
.pc .bar.on { display: block; } .pc .bar i { display: block; height: 100%; width: 0; background: var(--g); transition: width .9s linear; }
|
| 129 |
+
/* [1.2.3] a speed tile while an answer is in progress */
|
| 130 |
+
.lab .lv { display: none; font-weight: 400; color: #ff4d5e; margin-left: 6px; letter-spacing: .08em; }
|
| 131 |
+
.cell.live .lab .lv { display: inline; animation: lv-blink 1.2s ease-in-out infinite; }
|
| 132 |
+
@keyframes lv-blink { 50% { opacity: .35 } }
|
| 133 |
+
@media (prefers-reduced-motion: reduce) { .wire .pk, .pc.lit, .cell.live .lab .lv { animation: none !important; } }
|
| 134 |
.kv { display: grid; grid-template-columns: max-content 1fr; gap: 3px 12px; align-items: center; }
|
| 135 |
.kv .k { color: var(--muted); }
|
| 136 |
.kv code { font-family: var(--mono); font-size: 12px; background: var(--panel2); color: #a9b8ff; padding: 1px 6px; border-radius: 2px; }
|
|
|
|
| 238 |
<section class="page on" id="page-server">
|
| 239 |
<!-- [1.2] the telemetry strip: the last answers' speeds, memory, the NPU's load (or the context length) -->
|
| 240 |
<div class="strip">
|
| 241 |
+
<div class="cell main" id="c-gen"><span class="lab"><i>01</i><span>生成</span><b class="lv" id="lv-gen"></b></span><span class="num g off" id="t-gen">—</span><span class="unit">tok/s</span>
|
| 242 |
<svg class="spark" viewBox="0 0 100 22" preserveAspectRatio="none" aria-hidden="true"><defs><linearGradient id="sg" gradientUnits="userSpaceOnUse" x1="0" y1="0" x2="100" y2="0"><stop offset="0" stop-color="#ff4d5e"/><stop offset=".55" stop-color="#c44dff"/><stop offset="1" stop-color="#4d7dff"/></linearGradient></defs><polyline id="sp-gen" fill="none" stroke="url(#sg)" stroke-width="1.5" stroke-linejoin="round" vector-effect="non-scaling-stroke"/></svg></div>
|
| 243 |
+
<div class="cell" id="c-pre"><span class="lab"><i>02</i><span>プロンプト</span><b class="lv" id="lv-pre"></b></span><span class="num off" id="t-pre">—</span><span class="unit">tok/s</span>
|
| 244 |
<svg class="spark" viewBox="0 0 100 22" preserveAspectRatio="none" aria-hidden="true"><polyline id="sp-pre" fill="none" stroke="#8fb0ff" stroke-width="1.5" stroke-linejoin="round" vector-effect="non-scaling-stroke"/></svg></div>
|
| 245 |
<div class="cell" id="c-mem"><span class="lab"><i>03</i><span>メモリ</span></span><span class="num off" id="t-mem">—</span><span class="unit" id="t-memu">GB</span>
|
| 246 |
<span class="gauge" title="onwx-dash(サーバーと、NPU が使うメモリ)・ほかのアプリ・空き"><i class="u-onwx" id="u-onwx"></i><i class="u-oth" id="u-oth"></i></span>
|
|
|
|
| 276 |
<button class="primary" id="b-tp" onclick="tpToggle()">補助を開始</button>
|
| 277 |
</div>
|
| 278 |
<div class="link">
|
| 279 |
+
<div class="pc me" id="tp-mebox"><span class="lab" id="tp-me-role">—</span><b id="tp-me">—</b><span class="bar" id="tp-mebar"><i></i></span></div>
|
| 280 |
+
<div class="wire" id="tp-wire"><i class="pk f s0" style="--d:0s"></i><i class="pk f" style="--d:-.93s"></i><i class="pk f" style="--d:-1.86s"></i><i class="pk b" style="--d:-.45s"></i><i class="pk b" style="--d:-1.4s"></i><i class="pk b" style="--d:-2.33s"></i><span id="tp-addr">—</span></div>
|
| 281 |
+
<div class="pc" id="tp-peerbox"><span class="lab" id="tp-peer-role">—</span><b id="tp-peer">—</b><span class="bar" id="tp-peerbar"><i></i></span></div>
|
| 282 |
</div>
|
| 283 |
<div class="kv small" id="tp-modelrow">
|
| 284 |
<span class="k">モデル</span><span class="mono ellipsis" id="tp-model">—</span>
|
|
|
|
| 491 |
const cur = loadedNow && inst.some(m => m.key === sv.model) ? sv.model
|
| 492 |
: (pick.value && inst.some(m => m.key === pick.value)) ? pick.value
|
| 493 |
: (sv.model && inst.some(m => m.key === sv.model) ? sv.model : (inst.find(m => m.key === s.config.model) || inst[0] || {}).key);
|
| 494 |
+
const pickHtml = inst.map(m => `<option value="${esc(m.key)}"${m.key === cur ? " selected" : ""} title="${esc(m.path)}">${esc(m.title)}${m.key === sv.model ? T("(ロード済み)") : ""}</option>`).join("");
|
| 495 |
if (pickHtml !== lastPick) { pick.innerHTML = pickHtml; lastPick = pickHtml; } // rebuild only on change: a rebuild eats clicks
|
| 496 |
if (loadedNow && cur && pick.value !== cur) pick.value = cur;
|
| 497 |
const same = sv.model && pick.value === sv.model;
|
|
|
|
| 564 |
$("f-tp_enabled").checked = !!t.enabled; $("f-tp_role").value = t.role; $("f-tp_listener").value = t.listener;
|
| 565 |
$("f-tp_mode").value = t.ip === "auto" ? "auto" : tpRing(t.ip) ? "ring" : "manual"; $("f-tp_ip").value = t.ip === "auto" || tpRing(t.ip) ? "" : t.ip; $("f-tp_port").value = t.port; $("f-tp_streams").value = t.streams;
|
| 566 |
const opts = [`<option value="">${esc(T("自動(選んでいるモデルか、最初に見つかったもの)"))}</option>`]
|
| 567 |
+
.concat((s.tp.models || []).map(m => `<option value="${esc(m.key)}" title="${esc(m.path)}">${esc(m.title)}</option>`));
|
| 568 |
if (t.model && !(s.tp.models || []).some(m => m.key === t.model)) opts.push(`<option value="${esc(t.model)}">${esc(t.model)}</option>`);
|
| 569 |
const html = opts.join("");
|
| 570 |
if ($("f-tp_model").dataset.html !== html) { $("f-tp_model").innerHTML = html; $("f-tp_model").dataset.html = html; }
|
|
|
|
| 585 |
function setNum(id, text, on) { const e = $(id); if (e.textContent !== text) e.textContent = text; e.classList.toggle("off", !on); }
|
| 586 |
function renderStrip() {
|
| 587 |
const s = S, sv = s.server, live = sv.status === "running";
|
| 588 |
+
const sp = (live && s.speed) || [], lv = live && s.live || null, key = JSON.stringify(sp) + live + JSON.stringify(lv);
|
| 589 |
if (key !== lastSpeed) {
|
| 590 |
lastSpeed = key;
|
| 591 |
const gen = sp.map(x => x[0]).filter(x => x > 0), pre = sp.map(x => x[1]).filter(x => x > 0);
|
| 592 |
+
// [1.2.3] an answer in progress (asked once a second while this page shows): its speed now, the line moving
|
| 593 |
+
const g = !!lv && lv.phase === "gen", p = !!lv && lv.phase === "prompt";
|
| 594 |
+
const genNow = g && lv.tok_s > 0 ? gen.concat([lv.tok_s]) : gen;
|
| 595 |
+
const preNow = g && lv.pre > 0 ? pre.concat([lv.pre]) : p && lv.tok_s > 0 ? pre.concat([lv.tok_s]) : pre;
|
| 596 |
+
setNum("t-gen", g ? (lv.tok_s > 0 ? lv.tok_s.toFixed(1) : "…") : gen.length ? gen[gen.length - 1].toFixed(1) : "—", g || gen.length > 0);
|
| 597 |
+
setNum("t-pre", p ? (lv.tok_s > 0 ? Math.round(lv.tok_s).toLocaleString() : "…")
|
| 598 |
+
: preNow.length ? Math.round(preNow[preNow.length - 1]).toLocaleString() : "—", p || preNow.length > 0);
|
| 599 |
+
$("c-gen").classList.toggle("live", g); $("c-pre").classList.toggle("live", p);
|
| 600 |
+
$("lv-gen").textContent = g ? T("● 生成中 {0}", lv.tokens) : "";
|
| 601 |
+
$("lv-pre").textContent = p ? T("● 読み込み中 {0} 秒", Math.floor(lv.s)) : "";
|
| 602 |
+
spark("sp-gen", genNow); spark("sp-pre", preNow);
|
| 603 |
$("c-gen").title = $("c-pre").title = T("直近の回答の速さ。線は最近の回答の移り変わり");
|
| 604 |
}
|
| 605 |
// memory: onwx (the server and what the NPU holds for it) / other programs / free
|
|
|
|
| 710 |
: stg === "load" ? ["warn", T("メモリに読み込み中 {0}", pc(ld.gpu, ld.main_total))]
|
| 711 |
: stg === "helper" ? ["warn", T("接続しました")] : ["warn", T("補助 PC を待っています")];
|
| 712 |
$("tp-pill").className = "pill " + pl[0]; $("tp-pill").textContent = pl[1];
|
| 713 |
+
const busy = on && S.live && (S.live.phase === "gen" || S.live.phase === "prompt");
|
| 714 |
+
$("tp-wire").className = "wire" + (on || (tw && stg !== "wait") ? " on" : tw ? " wait" : "") + (busy ? " busy" : "") + (stg === "send" ? " send" : "");
|
| 715 |
+
tpBoxes(on || (tw && stg !== "wait"), tw && stg === "wait",
|
| 716 |
+
stg === "load" && ld.main_total ? ld.gpu / ld.main_total : null,
|
| 717 |
+
(stg === "send" || stg === "load") && ld.sent_total ? ld.sent / ld.sent_total : null);
|
| 718 |
$("tp-me-role").textContent = T("この PC ・ メイン"); $("tp-peer-role").textContent = T("補助 PC");
|
| 719 |
$("tp-me").textContent = on || tw ? (sv.title || "—") : "—";
|
| 720 |
$("tp-addr").textContent = tpShow(t.ip, t.port); $("tp-wire").title = tpAddr(t.listener === "main", t.ip, t.port);
|
|
|
|
| 732 |
const pl = {waiting: ["warn", T("メイン PC を待っています")], serving: ["ok", T("メイン PC とつながっています")],
|
| 733 |
failed: ["danger", T("補助が止まりました")], stopped: ["", T("補助は停止中")]}[h.status] || ["", h.status];
|
| 734 |
$("tp-pill").className = "pill " + pl[0]; $("tp-pill").textContent = pl[1];
|
| 735 |
+
$("tp-wire").className = "wire" + (h.status === "serving" ? " on" : h.status === "waiting" ? " wait" : "") +
|
| 736 |
+
(h.status === "serving" && h.loading && h.loading.stage === "recv" ? " recv" : "");
|
| 737 |
+
tpBoxes(h.status === "serving", h.status === "waiting",
|
| 738 |
+
h.status === "serving" && h.loading && h.loading.total ? h.loading.done / h.loading.total : null, null);
|
| 739 |
$("tp-me-role").textContent = T("この PC ・ 補助"); $("tp-peer-role").textContent = T("メイン PC");
|
| 740 |
const hl = h.status === "serving" && h.loading; // [1.2.2] the helper's half still loading
|
| 741 |
if (hl) { $("tp-pill").className = "pill warn";
|
|
|
|
| 756 |
: T("「補助を開始」を押すと、この PC のモデルを閉じてメイン PC を待ちます");
|
| 757 |
$("tp-msg").textContent = msg; $("tp-msg").className = "small grow " + (h.error ? "err" : "muted");
|
| 758 |
}
|
| 759 |
+
// [1.2.3] the two PC boxes: glowing while joined, dashed while waiting, a bar while a half loads (0..1 or null)
|
| 760 |
+
function tpBoxes(lit, wait, me, peer) {
|
| 761 |
+
for (const [box, bar, f] of [["tp-mebox", "tp-mebar", me], ["tp-peerbox", "tp-peerbar", peer]]) {
|
| 762 |
+
$(box).classList.toggle("lit", !!lit); $(box).classList.toggle("wait", !!wait);
|
| 763 |
+
$(bar).classList.toggle("on", f != null);
|
| 764 |
+
$(bar).firstChild.style.width = f != null ? Math.min(100, 100 * f).toFixed(1) + "%" : "0";
|
| 765 |
+
}
|
| 766 |
+
}
|
| 767 |
async function tpToggle() {
|
| 768 |
const h = S && S.tp && S.tp.helper, up = h && (h.status === "waiting" || h.status === "serving");
|
| 769 |
if (!up && S.server.model && !await ask(T("補助を開始すると、読み込まれているモデル({0})を閉じます。よろしいですか?", S.server.title || S.server.model))) return;
|
|
|
|
| 988 |
async function refresh() {
|
| 989 |
const n = ++REQ;
|
| 990 |
try {
|
| 991 |
+
// [1.2.3] the answer in progress only while the server page shows
|
| 992 |
+
const lv = $("page-server").classList.contains("on") && !document.hidden ? "&live=1" : "";
|
| 993 |
+
const r = await fetch("/api/state?token=" + encodeURIComponent(TOKEN) + lv, {cache: "no-store"});
|
| 994 |
if (r.status === 403) { $("env").textContent = T("この画面は onwx-dash のウィンドウかトレイから開いてください"); return; }
|
| 995 |
const j = await r.json();
|
| 996 |
if (n <= STALE) return; // (asked before the last save: shows the old values)
|