@infernetprotocol/cli 0.1.38 → 0.1.40
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/commands/inference.js +54 -5
- package/commands/register.js +25 -11
- package/lib/inference/petals-client.py.js +17 -0
- package/package.json +6 -6
package/commands/inference.js
CHANGED
|
@@ -32,6 +32,7 @@ Usage:
|
|
|
32
32
|
infernet inference serve --model <hf-id> [flags]
|
|
33
33
|
infernet inference status
|
|
34
34
|
infernet inference stop
|
|
35
|
+
infernet inference list
|
|
35
36
|
|
|
36
37
|
Flags (serve):
|
|
37
38
|
--model <hf-id> Required. e.g. meta-llama/Llama-3.1-70B-Instruct
|
|
@@ -126,13 +127,36 @@ async function cmdServe(args) {
|
|
|
126
127
|
prefix,
|
|
127
128
|
num_blocks: numBlocks,
|
|
128
129
|
pid: child.pid,
|
|
129
|
-
started_at: new Date().toISOString()
|
|
130
|
+
started_at: new Date().toISOString(),
|
|
131
|
+
peer_id: null
|
|
130
132
|
});
|
|
131
133
|
|
|
132
|
-
//
|
|
133
|
-
//
|
|
134
|
-
|
|
135
|
-
|
|
134
|
+
// Watch Petals' startup logs for the peer ID line — looks like
|
|
135
|
+
// "This server's peer ID: 12D3KooW..." (libp2p Ed25519). We
|
|
136
|
+
// capture it so `infernet register` can advertise it; the
|
|
137
|
+
// control plane needs the mapping to attribute per-layer CPR
|
|
138
|
+
// receipts to layer-contributing operators.
|
|
139
|
+
const tapForPeerId = (stream, sink) => {
|
|
140
|
+
let buf = "";
|
|
141
|
+
stream.on("data", async (chunk) => {
|
|
142
|
+
const txt = chunk.toString();
|
|
143
|
+
sink.write(txt);
|
|
144
|
+
buf += txt;
|
|
145
|
+
const m = buf.match(/peer ID[:\s]+([1-9A-HJ-NP-Za-km-z]{40,80})/i);
|
|
146
|
+
if (m) {
|
|
147
|
+
const state = (await readState()) ?? {};
|
|
148
|
+
if (!state.peer_id) {
|
|
149
|
+
state.peer_id = m[1];
|
|
150
|
+
await writeState(state);
|
|
151
|
+
process.stdout.write(`[infernet] captured petals peer id: ${m[1]}\n`);
|
|
152
|
+
}
|
|
153
|
+
buf = "";
|
|
154
|
+
}
|
|
155
|
+
if (buf.length > 4096) buf = buf.slice(-2048);
|
|
156
|
+
});
|
|
157
|
+
};
|
|
158
|
+
if (child.stdout) tapForPeerId(child.stdout, process.stdout);
|
|
159
|
+
if (child.stderr) tapForPeerId(child.stderr, process.stderr);
|
|
136
160
|
|
|
137
161
|
process.stdout.write(
|
|
138
162
|
`\n✓ Petals server pid=${child.pid} started.\n` +
|
|
@@ -171,6 +195,30 @@ async function cmdStatus() {
|
|
|
171
195
|
return 0;
|
|
172
196
|
}
|
|
173
197
|
|
|
198
|
+
async function cmdList() {
|
|
199
|
+
const cfg = await loadConfig().catch(() => ({}));
|
|
200
|
+
const controlPlaneUrl = cfg?.controlPlane?.url ?? process.env.NEXT_PUBLIC_APP_URL ?? "https://infernetprotocol.com";
|
|
201
|
+
const res = await fetch(`${controlPlaneUrl}/api/v1/petals/swarm`);
|
|
202
|
+
if (!res.ok) {
|
|
203
|
+
process.stderr.write(`error: HTTP ${res.status} from ${controlPlaneUrl}\n`);
|
|
204
|
+
return 1;
|
|
205
|
+
}
|
|
206
|
+
const { data } = await res.json();
|
|
207
|
+
if (!data?.models?.length) {
|
|
208
|
+
process.stdout.write(`(no Petals swarms active right now)\n`);
|
|
209
|
+
process.stdout.write(`Be the first — \`infernet inference serve --backend petals --model <hf-id>\`\n`);
|
|
210
|
+
return 0;
|
|
211
|
+
}
|
|
212
|
+
process.stdout.write(`\nPetals swarms — ${data.total_models} models · ${data.total_nodes} nodes serving\n\n`);
|
|
213
|
+
process.stdout.write(` ${"NODES".padEnd(7)} MODEL\n`);
|
|
214
|
+
process.stdout.write(` ${"─────".padEnd(7)} ─────\n`);
|
|
215
|
+
for (const m of data.models) {
|
|
216
|
+
process.stdout.write(` ${String(m.node_count).padEnd(7)} ${m.model}\n`);
|
|
217
|
+
}
|
|
218
|
+
process.stdout.write(`\nQuery as a user: tick \"Distribute across all nodes\" on /chat and pick a model above.\n\n`);
|
|
219
|
+
return 0;
|
|
220
|
+
}
|
|
221
|
+
|
|
174
222
|
async function cmdStop() {
|
|
175
223
|
const state = await readState();
|
|
176
224
|
if (!state?.pid) {
|
|
@@ -197,6 +245,7 @@ export default async function inference(args) {
|
|
|
197
245
|
case "serve": case "start": return cmdServe(args);
|
|
198
246
|
case "status": return cmdStatus();
|
|
199
247
|
case "stop": return cmdStop();
|
|
248
|
+
case "list": case "ls": return cmdList();
|
|
200
249
|
default:
|
|
201
250
|
process.stderr.write(sub ? `unknown subcommand: ${sub}\n\n` : "error: missing subcommand\n\n");
|
|
202
251
|
process.stderr.write(HELP);
|
package/commands/register.js
CHANGED
|
@@ -208,27 +208,41 @@ export async function gatherCoarseSpecs() {
|
|
|
208
208
|
// IPIP-0031: models this node is contributing to a Petals swarm.
|
|
209
209
|
// Empty array if the operator hasn't run `infernet inference serve`.
|
|
210
210
|
// Read from local state file written by inference.js.
|
|
211
|
-
petals_models: await readPetalsModels()
|
|
211
|
+
petals_models: await readPetalsModels(),
|
|
212
|
+
// IPIP-0031 §per-layer-attribution: libp2p peer ID Petals
|
|
213
|
+
// assigned this server. The control plane uses this to map
|
|
214
|
+
// chosen_servers from a client's routing event back to the
|
|
215
|
+
// corresponding provider rows for CPR receipt distribution.
|
|
216
|
+
petals_peer_id: await readPetalsPeerId()
|
|
212
217
|
};
|
|
213
218
|
}
|
|
214
219
|
|
|
215
220
|
async function readPetalsModels() {
|
|
221
|
+
const state = await readPetalsState();
|
|
222
|
+
if (!state) return [];
|
|
223
|
+
if (state.backend === 'petals' && typeof state.model === 'string') {
|
|
224
|
+
return [state.model];
|
|
225
|
+
}
|
|
226
|
+
return [];
|
|
227
|
+
}
|
|
228
|
+
|
|
229
|
+
async function readPetalsPeerId() {
|
|
230
|
+
const state = await readPetalsState();
|
|
231
|
+
return (state && typeof state.peer_id === 'string') ? state.peer_id : null;
|
|
232
|
+
}
|
|
233
|
+
|
|
234
|
+
async function readPetalsState() {
|
|
216
235
|
try {
|
|
217
236
|
const fsp = await import('node:fs/promises');
|
|
218
237
|
const path = await import('node:path');
|
|
219
238
|
const statePath = path.join(process.env.HOME ?? '/tmp', '.infernet', 'inference', 'state.json');
|
|
220
239
|
const raw = await fsp.readFile(statePath, 'utf8');
|
|
221
240
|
const state = JSON.parse(raw);
|
|
222
|
-
|
|
223
|
-
|
|
224
|
-
|
|
225
|
-
|
|
226
|
-
|
|
227
|
-
if (state.backend === 'petals' && typeof state.model === 'string') {
|
|
228
|
-
return [state.model];
|
|
229
|
-
}
|
|
230
|
-
return [];
|
|
231
|
-
} catch { return []; }
|
|
241
|
+
if (!state.pid) return null;
|
|
242
|
+
try { process.kill(state.pid, 0); }
|
|
243
|
+
catch { return null; }
|
|
244
|
+
return state;
|
|
245
|
+
} catch { return null; }
|
|
232
246
|
}
|
|
233
247
|
|
|
234
248
|
export default async function register(args, ctx) {
|
|
@@ -72,6 +72,23 @@ def main():
|
|
|
72
72
|
full_text += token_text
|
|
73
73
|
emit({"event": "token", "text": token_text})
|
|
74
74
|
|
|
75
|
+
# IPIP-0031 per-layer attribution: after the first generate the
|
|
76
|
+
# session has resolved which DHT peers serve which transformer
|
|
77
|
+
# blocks. Emit that mapping so the control plane can split the
|
|
78
|
+
# CPR receipt across layer-contributing operators by block share.
|
|
79
|
+
try:
|
|
80
|
+
chosen = []
|
|
81
|
+
for s in getattr(session, "chosen_servers", []) or []:
|
|
82
|
+
chosen.append({
|
|
83
|
+
"peer_id": str(getattr(s, "peer_id", "")),
|
|
84
|
+
"start_block": int(getattr(s, "start", getattr(s, "start_block", 0))),
|
|
85
|
+
"end_block": int(getattr(s, "end", getattr(s, "end_block", 0))),
|
|
86
|
+
})
|
|
87
|
+
if chosen:
|
|
88
|
+
emit({"event": "routing", "peers": chosen})
|
|
89
|
+
except Exception as e:
|
|
90
|
+
emit({"event": "log", "warn": f"chosen_servers extract failed: {e}"})
|
|
91
|
+
|
|
75
92
|
for _ in range(max_tokens - 1):
|
|
76
93
|
outputs = model_obj.generate(
|
|
77
94
|
outputs,
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@infernetprotocol/cli",
|
|
3
|
-
"version": "0.1.
|
|
3
|
+
"version": "0.1.40",
|
|
4
4
|
"description": "GPU-node CLI / daemon for the Infernet Protocol — register a GPU server with a Supabase control plane and start earning.",
|
|
5
5
|
"license": "MIT",
|
|
6
6
|
"type": "module",
|
|
@@ -30,11 +30,11 @@
|
|
|
30
30
|
"ora": "^9.4.0",
|
|
31
31
|
"react": "^17.0.2",
|
|
32
32
|
"react-blessed": "^0.7.2",
|
|
33
|
-
"@infernetprotocol/auth": "0.1.
|
|
34
|
-
"@infernetprotocol/
|
|
35
|
-
"@infernetprotocol/
|
|
36
|
-
"@infernetprotocol/
|
|
37
|
-
"@infernetprotocol/gpu": "0.1.
|
|
33
|
+
"@infernetprotocol/auth": "0.1.40",
|
|
34
|
+
"@infernetprotocol/deploy-providers": "0.1.40",
|
|
35
|
+
"@infernetprotocol/engine": "0.1.40",
|
|
36
|
+
"@infernetprotocol/config": "0.1.40",
|
|
37
|
+
"@infernetprotocol/gpu": "0.1.40"
|
|
38
38
|
},
|
|
39
39
|
"repository": {
|
|
40
40
|
"type": "git",
|