@infernetprotocol/cli 0.1.94 → 0.1.95

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/setup.js CHANGED
@@ -218,7 +218,12 @@ async function startOllama({ yes }) {
218
218
  // --user service or simply lack passwordless sudo.
219
219
  const uid = process.getuid?.() ?? null;
220
220
  const userBus = uid != null && fsSync.existsSync(`/run/user/${uid}/bus`);
221
- const NOHUP = "(nohup ollama serve > /tmp/ollama.log 2>&1 < /dev/null &)";
221
+ // Flash attention + q8_0 KV cache unless the operator set them: halves
222
+ // the KV cache, so long-context models fit the GPU (see install.sh
223
+ // ensure_ollama_memory_env, which does the same for ollama.service).
224
+ const NOHUP = "(OLLAMA_FLASH_ATTENTION=${OLLAMA_FLASH_ATTENTION:-1}"
225
+ + " OLLAMA_KV_CACHE_TYPE=${OLLAMA_KV_CACHE_TYPE:-q8_0}"
226
+ + " nohup ollama serve > /tmp/ollama.log 2>&1 < /dev/null &)";
222
227
  let cmd;
223
228
  if (isMac) {
224
229
  cmd = `open -a Ollama || ${NOHUP}`;
@@ -157,6 +157,11 @@ async function defaultEngine() {
157
157
  if (Number.isFinite(eng.ollama_num_thread) && eng.ollama_num_thread > 0) {
158
158
  opts.numThread = eng.ollama_num_thread;
159
159
  }
160
+ // Optional override for the per-request context cap (default 32768,
161
+ // INFERNET_OLLAMA_MAX_CTX; 0 = no cap). See DEFAULT_MAX_CTX in the engine.
162
+ if (Number.isFinite(eng.ollama_max_ctx) && eng.ollama_max_ctx >= 0) {
163
+ opts.maxCtx = eng.ollama_max_ctx;
164
+ }
160
165
  const cacheKey = `default:${eng.backend ?? 'auto'}`;
161
166
  const engine = await buildEngine(cacheKey, opts);
162
167
  if (engine.kind === 'stub' && !stubExplicitlyChosen(config)) {
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@infernetprotocol/cli",
3
- "version": "0.1.94",
3
+ "version": "0.1.95",
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",
@@ -32,14 +32,14 @@
32
32
  "ora": "^9.4.1",
33
33
  "react": "^17.0.2",
34
34
  "react-blessed": "^0.7.2",
35
- "@infernetprotocol/auth": "0.1.94",
36
- "@infernetprotocol/config": "0.1.94",
37
- "@infernetprotocol/engine": "0.1.94",
38
- "@infernetprotocol/deploy-providers": "0.1.94",
39
- "@infernetprotocol/gpu": "0.1.94",
40
- "@infernetprotocol/liquid-agent": "0.1.94",
41
- "@infernetprotocol/rpc-adapter": "0.1.94",
42
- "@infernetprotocol/training": "0.1.94"
35
+ "@infernetprotocol/auth": "0.1.95",
36
+ "@infernetprotocol/config": "0.1.95",
37
+ "@infernetprotocol/deploy-providers": "0.1.95",
38
+ "@infernetprotocol/engine": "0.1.95",
39
+ "@infernetprotocol/gpu": "0.1.95",
40
+ "@infernetprotocol/liquid-agent": "0.1.95",
41
+ "@infernetprotocol/rpc-adapter": "0.1.95",
42
+ "@infernetprotocol/training": "0.1.95"
43
43
  },
44
44
  "repository": {
45
45
  "type": "git",