@atomic-chat/core 0.5.1

package/dist/runtime/llamacpp/args.js

@@ -0,0 +1,379 @@
+/**
+ * llama-server argv builder. Port of
+ * `src-tauri/plugins/tauri-plugin-llamacpp-upstream/src/args.rs` (ArgumentBuilder).
+ *
+ * Pure: takes a full `LlamacppConfig` and load inputs, returns argv + warnings. Emission order and
+ * every gate are pinned by `test/contract/args.test.ts` against fixtures emitted from the Rust
+ * implementation (comparator `argv-exact`).
+ *
+ * Provider dependence (PLAN.md §8.2 "Divergences"): only the cache-type whitelist differs —
+ * `llamacpp-upstream` never emits fork-only `turbo*` types, `llamacpp` (TurboQuant) allows them when
+ * the installed version is a fork tag. Everything else is shared.
+ */
+import { AtomicCoreError } from '../../contracts/index.js';
+import { f32Differs, formatRustF32 } from '../../util/index.js';
+export { f32Differs, formatRustF32 };
+/** Minimum llama.cpp build that turned `--flash-attn` into a string argument (auto|on|off). */
+export const FLASH_ATTN_STRING_ARG_MIN_BUILD = 6325;
+/** First upstream release containing Qwen built-in MTP (`--spec-type draft-mtp`). */
+export const MTP_MIN_BUILD = 9180;
+/** First upstream release containing the Gemma 4 separate-draft MTP merge. */
+export const GEMMA_MTP_MIN_BUILD = 9553;
+/** First upstream release containing `--reasoning-preserve`. */
+export const REASONING_PRESERVE_MIN_BUILD = 9837;
+/** First upstream release where reasoning preservation is on by default (silence means "on"). */
+export const REASONING_PRESERVE_DEFAULT_ON_MIN_BUILD = 10762;
+/** Cache types supported by stock ggml-org builds. */
+export const STANDARD_CACHE_TYPES = [
+    'f32',
+    'f16',
+    'bf16',
+    'q8_0',
+    'q4_0',
+    'q4_1',
+    'iq4_nl',
+    'q5_0',
+    'q5_1',
+];
+/** Fields with `#[serde(default)]` in Rust; every other `LlamacppConfig` field is required. */
+/**
+ * Fields Rust deserialisation fills in when the app leaves them out. The two string fields are not
+ * engine settings — the extension carries them per model — so a core that loads a model the app has
+ * never configured must supply `String::default()` itself, or the builder emits a flag with no value.
+ */
+export const LLAMACPP_CONFIG_SERDE_DEFAULTS = {
+    chat_template: '',
+    override_tensor_buffer_t: '',
+    parallel: 1,
+    concurrent_mode: false,
+    concurrent_slots: 8,
+    expose_metrics: false,
+    mtp: false,
+    mtp_draft_path: '',
+    dflash: false,
+    dflash_spec_supported: false,
+    dflash_draft_path: '',
+    dflash_n_max: 0,
+    reasoning_preserve: false,
+    extra_args: '',
+};
+/**
+ * Fill the serde-default fields the way Rust deserialisation would. A key present with the value
+ * `undefined` counts as absent — in JSON it would simply not be there, and letting it through would
+ * overwrite a default with nothing.
+ */
+export function withLlamacppDefaults(input) {
+    const provided = Object.fromEntries(Object.entries(input).filter(([, v]) => v !== undefined));
+    return { ...LLAMACPP_CONFIG_SERDE_DEFAULTS, ...provided };
+}
+/** `"<version>/<backend>"` → parts. BOM stripped, both sides trimmed. Throws on a missing `/`. */
+export function parseVersionBackend(versionBackend) {
+    const cleaned = versionBackend.replaceAll('', '');
+    const slash = cleaned.indexOf('/');
+    if (slash < 0) {
+        throw new AtomicCoreError('INVALID_ARGUMENT', 'Invalid configuration argument provided.', 'Invalid version_backend format');
+    }
+    return { version: cleaned.slice(0, slash).trim(), backend: cleaned.slice(slash + 1).trim() };
+}
+/** `"b6325"` → 6325, `"b10018-1.3.0"` → 10018, `"v1.0"` → undefined. */
+export function parseBuildNumber(version) {
+    if (!version.startsWith('b'))
+        return undefined;
+    const build = version.slice(1).split('-')[0] ?? '';
+    if (!/^\d+$/.test(build))
+        return undefined;
+    const n = Number(build);
+    return n <= 0xffff_ffff ? n : undefined;
+}
+/** TurboQuant release train: `turboquant-*` or the unified `b<build>-<x>.<y>.<z>` tag. */
+export function isTurboquantVersion(version) {
+    if (version.startsWith('turboquant-'))
+        return true;
+    if (!version.startsWith('b'))
+        return false;
+    const rest = version.slice(1);
+    const dash = rest.indexOf('-');
+    if (dash < 0)
+        return false;
+    const build = rest.slice(0, dash);
+    const semver = rest.slice(dash + 1);
+    const isNumber = (s) => /^\d+$/.test(s);
+    if (!isNumber(build))
+        return false;
+    const parts = semver.split('.');
+    return parts.length === 3 && parts.every(isNumber);
+}
+/**
+ * POSIX-ish splitter for user-supplied extra arguments. Quotes group; backslash escapes the active
+ * quote or a backslash inside quotes, and whitespace/quotes/backslash outside. An unterminated quote
+ * is an error (the caller drops the whole string).
+ */
+export function parseExtraArgs(value) {
+    const args = [];
+    let current = '';
+    let quote = null;
+    let started = false;
+    const chars = Array.from(value);
+    for (let i = 0; i < chars.length; i++) {
+        const ch = chars[i];
+        const next = chars[i + 1];
+        if (quote !== null) {
+            if (ch === quote)
+                quote = null;
+            else if (ch === '\\' && (next === quote || next === '\\')) {
+                current += next;
+                i++;
+            }
+            else
+                current += ch;
+            continue;
+        }
+        if (ch === "'" || ch === '"') {
+            quote = ch;
+            started = true;
+        }
+        else if (ch === '\\' &&
+            next !== undefined &&
+            (/\s/.test(next) || next === "'" || next === '"' || next === '\\')) {
+            current += next;
+            i++;
+            started = true;
+        }
+        else if (/\s/.test(ch)) {
+            if (started) {
+                args.push(current);
+                current = '';
+                started = false;
+            }
+        }
+        else {
+            current += ch;
+            started = true;
+        }
+    }
+    if (quote !== null)
+        throw new Error(`unterminated ${quote} quote`);
+    if (started)
+        args.push(current);
+    return args;
+}
+/** Build the plan (argv + warnings). Throws `INVALID_ARGUMENT` for a malformed `version_backend`. */
+export function planLlamaArgs(config, input) {
+    const { version, backend } = parseVersionBackend(config.version_backend);
+    const cfg = { ...config };
+    const argv = [];
+    const warnings = [];
+    const build = parseBuildNumber(version);
+    const turboquant = isTurboquantVersion(version);
+    const ik = backend.startsWith('ik');
+    const push = (...items) => argv.push(...items);
+    // Pre-pass: concurrent mode, then Vulkan safety override.
+    if (cfg.concurrent_mode) {
+        cfg.parallel = Math.max(cfg.concurrent_slots, 2);
+        cfg.cont_batching = true;
+        cfg.expose_metrics = true;
+    }
+    // The TurboQuant plugin has neither the Vulkan override nor speculative decoding. The override
+    // matters beyond flash attention: `--cache-type-v` is only emitted with flash attention on, so
+    // forcing `auto` off on a TurboQuant Vulkan build would silently drop its default turbo3 V cache.
+    if (input.provider === 'llamacpp') {
+        cfg.mtp = false;
+        cfg.dflash = false;
+    }
+    if (input.provider !== 'llamacpp' && backend.includes('vulkan') && cfg.flash_attn === 'auto') {
+        warnings.push(`Vulkan backend (${backend}): overriding flash_attn auto→off for stability (ATO-244)`);
+        cfg.flash_attn = 'off';
+    }
+    const sanitizeCacheType = (value) => {
+        if (STANDARD_CACHE_TYPES.includes(value))
+            return value;
+        if (input.provider === 'llamacpp' && turboquant)
+            return value;
+        warnings.push(`Cache type '${value}' is not supported by non-turboquant backend ${version}/${backend}; falling back to q8_0`);
+        return 'q8_0';
+    };
+    if (!ik)
+        push('--no-webui');
+    push('--jinja');
+    push('-m', input.modelPath);
+    if (cfg.cpu_moe)
+        push('--cpu-moe');
+    if (cfg.n_cpu_moe > 0)
+        push('--n-cpu-moe', String(cfg.n_cpu_moe));
+    if (cfg.override_tensor_buffer_t)
+        push('--override-tensor', cfg.override_tensor_buffer_t);
+    if (input.mmprojPath) {
+        push('--mmproj', input.mmprojPath);
+        if (!cfg.offload_mmproj)
+            push('--no-mmproj-offload');
+    }
+    push('-a', input.modelId);
+    push('--port', String(input.port));
+    if (cfg.chat_template)
+        push('--chat-template', cfg.chat_template);
+    const ngl = cfg.n_gpu_layers >= 0 && cfg.n_gpu_layers !== 100 ? cfg.n_gpu_layers : -1;
+    push('-ngl', String(ngl));
+    if (cfg.threads > 0)
+        push('--threads', String(cfg.threads));
+    if (cfg.threads_batch > 0)
+        push('--threads-batch', String(cfg.threads_batch));
+    if (cfg.batch_size > 0 && cfg.batch_size !== 2048)
+        push('--batch-size', String(cfg.batch_size));
+    if (cfg.ubatch_size > 0 && cfg.ubatch_size !== 512)
+        push('--ubatch-size', String(cfg.ubatch_size));
+    if (cfg.device !== '')
+        push('--device', cfg.device);
+    if (cfg.split_mode !== '' && cfg.split_mode !== 'layer')
+        push('--split-mode', cfg.split_mode);
+    if (cfg.main_gpu !== 0)
+        push('--main-gpu', String(cfg.main_gpu));
+    // Flash attention.
+    if (ik) {
+        if (cfg.flash_attn === 'on')
+            push('-fa');
+    }
+    else {
+        const stringArg = turboquant || (build !== undefined && build >= FLASH_ATTN_STRING_ARG_MIN_BUILD);
+        if (stringArg) {
+            if (cfg.flash_attn === 'auto' || cfg.flash_attn === 'on' || cfg.flash_attn === 'off') {
+                push('--flash-attn', cfg.flash_attn);
+            }
+        }
+        else if (cfg.flash_attn === 'on') {
+            push('--flash-attn');
+        }
+    }
+    if (cfg.ctx_shift)
+        push('--context-shift');
+    if (cfg.cont_batching)
+        push('--cont-batching');
+    if (cfg.no_mmap)
+        push('--no-mmap');
+    if (cfg.mlock)
+        push('--mlock');
+    if (cfg.no_kv_offload)
+        push('--no-kv-offload');
+    if (cfg.parallel > 0) {
+        push('--parallel', String(cfg.parallel));
+        if (cfg.parallel === 1)
+            push('-kvu'); // ggml-org/llama.cpp#17450
+    }
+    // Speculative decoding: DFlash wins when both are set.
+    const addMtp = () => {
+        if (!cfg.mtp || input.isEmbedding)
+            return;
+        if (cfg.mtp_draft_path !== '') {
+            if (!(build !== undefined && build >= GEMMA_MTP_MIN_BUILD)) {
+                warnings.push(`Gemma 4 MTP requested but backend build ${version}/${backend} predates the Gemma MTP merge (b${GEMMA_MTP_MIN_BUILD}); skipping`);
+                return;
+            }
+            const k = cfg.cache_type_k;
+            const v = cfg.cache_type_v;
+            if ((k !== '' && k !== 'f16') || (v !== '' && v !== 'f16')) {
+                warnings.push(`Gemma 4 MTP is enabled with quantized KV cache (k=${k || 'f16'}, v=${v || 'f16'}); draft acceptance may drop`);
+            }
+            push('--model-draft', cfg.mtp_draft_path, '--spec-type', 'draft-mtp', '--spec-draft-n-max', '4');
+            return;
+        }
+        if (!(build !== undefined && build >= MTP_MIN_BUILD)) {
+            warnings.push(`MTP requested but backend build ${version}/${backend} predates upstream MTP merge (b${MTP_MIN_BUILD}); skipping`);
+            return;
+        }
+        push('--spec-type', 'draft-mtp', '--spec-draft-n-max', '2');
+    };
+    const addDflash = () => {
+        if (!cfg.dflash || input.isEmbedding)
+            return;
+        if (cfg.dflash_draft_path === '') {
+            warnings.push('DFlash requested but no draft GGUF path is set; skipping');
+            return;
+        }
+        if (!cfg.dflash_spec_supported) {
+            warnings.push(`DFlash requested for backend ${version}/${backend} but this llama.cpp build does not advertise --spec-type draft-dflash; loading without DFlash`);
+            return;
+        }
+        push('--model-draft', cfg.dflash_draft_path, '--spec-type', 'draft-dflash', '--spec-draft-n-max', String(cfg.dflash_n_max > 0 ? cfg.dflash_n_max : 15));
+    };
+    if (cfg.dflash && cfg.mtp) {
+        warnings.push('Both MTP and DFlash are enabled; applying DFlash only');
+        addDflash();
+    }
+    else {
+        addMtp();
+        addDflash();
+    }
+    if (cfg.expose_metrics)
+        push('--metrics');
+    // Reasoning preservation.
+    if (!input.isEmbedding) {
+        if (!cfg.reasoning_preserve) {
+            if (build !== undefined && build >= REASONING_PRESERVE_DEFAULT_ON_MIN_BUILD)
+                push('--no-reasoning-preserve');
+        }
+        else if (build !== undefined && build >= REASONING_PRESERVE_MIN_BUILD) {
+            push('--reasoning-preserve');
+        }
+        else {
+            warnings.push(`Reasoning preservation requested but backend build ${version}/${backend} predates upstream support (b${REASONING_PRESERVE_MIN_BUILD}); skipping`);
+        }
+    }
+    if (input.isEmbedding) {
+        push('--embedding', '--pooling', 'mean');
+    }
+    else {
+        if (cfg.ctx_size > 0 && !cfg.fit)
+            push('--ctx-size', String(cfg.ctx_size));
+        if (cfg.n_predict > 0)
+            push('--n-predict', String(cfg.n_predict));
+        if (cfg.cache_type_k !== '' && cfg.cache_type_k !== 'f16') {
+            const safeK = sanitizeCacheType(cfg.cache_type_k);
+            if (safeK !== 'f16')
+                push('--cache-type-k', safeK);
+        }
+        if (cfg.flash_attn !== 'off' &&
+            cfg.cache_type_v !== '' &&
+            cfg.cache_type_v !== 'f16' &&
+            cfg.cache_type_v !== 'f32') {
+            const safeV = sanitizeCacheType(cfg.cache_type_v);
+            if (safeV !== 'f16' && safeV !== 'f32')
+                push('--cache-type-v', safeV);
+        }
+        if (f32Differs(cfg.defrag_thold, 0.1))
+            push('--defrag-thold', formatRustF32(cfg.defrag_thold));
+        if (cfg.rope_scaling !== '' && cfg.rope_scaling !== 'none')
+            push('--rope-scaling', cfg.rope_scaling);
+        if (f32Differs(cfg.rope_scale, 1.0))
+            push('--rope-scale', formatRustF32(cfg.rope_scale));
+        if (Math.fround(cfg.rope_freq_base) !== 0)
+            push('--rope-freq-base', formatRustF32(cfg.rope_freq_base));
+        if (f32Differs(cfg.rope_freq_scale, 1.0))
+            push('--rope-freq-scale', formatRustF32(cfg.rope_freq_scale));
+    }
+    if (!ik) {
+        push('--fit', cfg.fit ? 'on' : 'off');
+        if (cfg.fit) {
+            if (cfg.fit_ctx !== '' && cfg.fit_ctx !== '4096')
+                push('--fit-ctx', cfg.fit_ctx);
+            if (cfg.fit_target !== '' && cfg.fit_target !== '1024')
+                push('--fit-target', cfg.fit_target);
+        }
+    }
+    try {
+        push(...parseExtraArgs(cfg.extra_args));
+    }
+    catch (error) {
+        warnings.push(`Ignoring invalid extra llama-server arguments: ${error.message}`);
+    }
+    // A config field that arrives as null/undefined would otherwise reach the process as the string
+    // "null"; Rust's types made that impossible, so the port checks it explicitly.
+    const bad = argv.findIndex((a) => typeof a !== 'string');
+    if (bad >= 0) {
+        throw new AtomicCoreError('INVALID_ARGUMENT', 'Invalid configuration argument provided.', `argument ${bad} after "${argv[bad - 1] ?? ''}" is ${String(argv[bad])}`);
+    }
+    return { argv, warnings, version, backend };
+}
+/** Convenience: argv only. */
+export function buildLlamaArgs(config, input) {
+    return planLlamaArgs(config, input).argv;
+}
+//# sourceMappingURL=args.js.map

package/dist/runtime/llamacpp/args.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"args.js","sourceRoot":"","sources":["../../../src/runtime/llamacpp/args.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,EAAE,eAAe,EAAE,MAAM,0BAA0B,CAAA;AAE1D,OAAO,EAAE,UAAU,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAA;AAE/D,OAAO,EAAE,UAAU,EAAE,aAAa,EAAE,CAAA;AAEpC,+FAA+F;AAC/F,MAAM,CAAC,MAAM,+BAA+B,GAAG,IAAI,CAAA;AACnD,qFAAqF;AACrF,MAAM,CAAC,MAAM,aAAa,GAAG,IAAI,CAAA;AACjC,8EAA8E;AAC9E,MAAM,CAAC,MAAM,mBAAmB,GAAG,IAAI,CAAA;AACvC,gEAAgE;AAChE,MAAM,CAAC,MAAM,4BAA4B,GAAG,IAAI,CAAA;AAChD,iGAAiG;AACjG,MAAM,CAAC,MAAM,uCAAuC,GAAG,KAAK,CAAA;AAE5D,sDAAsD;AACtD,MAAM,CAAC,MAAM,oBAAoB,GAAsB;IACrD,KAAK;IACL,KAAK;IACL,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,QAAQ;IACR,MAAM;IACN,MAAM;CACP,CAAA;AAED,+FAA+F;AAC/F;;;;GAIG;AACH,MAAM,CAAC,MAAM,8BAA8B,GAAG;IAC5C,aAAa,EAAE,EAAE;IACjB,wBAAwB,EAAE,EAAE;IAC5B,QAAQ,EAAE,CAAC;IACX,eAAe,EAAE,KAAK;IACtB,gBAAgB,EAAE,CAAC;IACnB,cAAc,EAAE,KAAK;IACrB,GAAG,EAAE,KAAK;IACV,cAAc,EAAE,EAAE;IAClB,MAAM,EAAE,KAAK;IACb,qBAAqB,EAAE,KAAK;IAC5B,iBAAiB,EAAE,EAAE;IACrB,YAAY,EAAE,CAAC;IACf,kBAAkB,EAAE,KAAK;IACzB,UAAU,EAAE,EAAE;CAC4B,CAAA;AAK5C;;;;GAIG;AACH,MAAM,UAAU,oBAAoB,CAAC,KAA0B;IAC7D,MAAM,QAAQ,GAAG,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,KAAK,SAAS,CAAC,CAAC,CAAA;IAC7F,OAAO,EAAE,GAAG,8BAA8B,EAAE,GAAG,QAAQ,EAAoB,CAAA;AAC7E,CAAC;AAmBD,kGAAkG;AAClG,MAAM,UAAU,mBAAmB,CAAC,cAAsB;IACxD,MAAM,OAAO,GAAG,cAAc,CAAC,UAAU,CAAC,GAAG,EAAE,EAAE,CAAC,CAAA;IAClD,MAAM,KAAK,GAAG,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,CAAA;IAClC,IAAI,KAAK,GAAG,CAAC,EAAE,CAAC;QACd,MAAM,IAAI,eAAe,CACvB,kBAAkB,EAClB,0CAA0C,EAC1C,gCAAgC,CACjC,CAAA;IACH,CAAC;IACD,OAAO,EAAE,OAAO,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,IAAI,EAAE,EAAE,OAAO,EAAE,OAAO,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,EAAE,CAAA;AAC9F,CAAC;AAED,wEAAwE;AACxE,MAAM,UAAU,gBAAgB,CAAC,OAAe;IAC9C,IAAI,CAAC,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC;QAAE,OAAO,SAAS,CAAA;IAC9C,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAA;IAClD,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC;QAAE,OAAO,SAAS,CAAA;IAC1C,MAAM,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,CAAA;IACvB,OAAO,CAAC,IAAI,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAA;AACzC,CAAC;AAED,0FAA0F;AAC1F,MAAM,UAAU,mBAAmB,CAAC,OAAe;IACjD,IAAI,OAAO,CAAC,UAAU,CAAC,aAAa,CAAC;QAAE,OAAO,IAAI,CAAA;IAClD,IAAI,CAAC,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC;QAAE,OAAO,KAAK,CAAA;IAC1C,MAAM,IAAI,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAA;IAC7B,MAAM,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAA;IAC9B,IAAI,IAAI,GAAG,CAAC;QAAE,OAAO,KAAK,CAAA;IAC1B,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,CAAA;IACjC,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,GAAG,CAAC,CAAC,CAAA;IACnC,MAAM,QAAQ,GAAG,CAAC,CAAS,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAA;IAC/C,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC;QAAE,OAAO,KAAK,CAAA;IAClC,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,CAAA;IAC/B,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC,IAAI,KAAK,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAA;AACpD,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,cAAc,CAAC,KAAa;IAC1C,MAAM,IAAI,GAAa,EAAE,CAAA;IACzB,IAAI,OAAO,GAAG,EAAE,CAAA;IAChB,IAAI,KAAK,GAAkB,IAAI,CAAA;IAC/B,IAAI,OAAO,GAAG,KAAK,CAAA;IACnB,MAAM,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAA;IAC/B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACtC,MAAM,EAAE,GAAG,KAAK,CAAC,CAAC,CAAW,CAAA;QAC7B,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,CAAA;QACzB,IAAI,KAAK,KAAK,IAAI,EAAE,CAAC;YACnB,IAAI,EAAE,KAAK,KAAK;gBAAE,KAAK,GAAG,IAAI,CAAA;iBACzB,IAAI,EAAE,KAAK,IAAI,IAAI,CAAC,IAAI,KAAK,KAAK,IAAI,IAAI,KAAK,IAAI,CAAC,EAAE,CAAC;gBAC1D,OAAO,IAAI,IAAI,CAAA;gBACf,CAAC,EAAE,CAAA;YACL,CAAC;;gBAAM,OAAO,IAAI,EAAE,CAAA;YACpB,SAAQ;QACV,CAAC;QACD,IAAI,EAAE,KAAK,GAAG,IAAI,EAAE,KAAK,GAAG,EAAE,CAAC;YAC7B,KAAK,GAAG,EAAE,CAAA;YACV,OAAO,GAAG,IAAI,CAAA;QAChB,CAAC;aAAM,IACL,EAAE,KAAK,IAAI;YACX,IAAI,KAAK,SAAS;YAClB,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,IAAI,KAAK,GAAG,IAAI,IAAI,KAAK,GAAG,IAAI,IAAI,KAAK,IAAI,CAAC,EAClE,CAAC;YACD,OAAO,IAAI,IAAI,CAAA;YACf,CAAC,EAAE,CAAA;YACH,OAAO,GAAG,IAAI,CAAA;QAChB,CAAC;aAAM,IAAI,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,CAAC;YACzB,IAAI,OAAO,EAAE,CAAC;gBACZ,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,CAAA;gBAClB,OAAO,GAAG,EAAE,CAAA;gBACZ,OAAO,GAAG,KAAK,CAAA;YACjB,CAAC;QACH,CAAC;aAAM,CAAC;YACN,OAAO,IAAI,EAAE,CAAA;YACb,OAAO,GAAG,IAAI,CAAA;QAChB,CAAC;IACH,CAAC;IACD,IAAI,KAAK,KAAK,IAAI;QAAE,MAAM,IAAI,KAAK,CAAC,gBAAgB,KAAK,QAAQ,CAAC,CAAA;IAClE,IAAI,OAAO;QAAE,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,CAAA;IAC/B,OAAO,IAAI,CAAA;AACb,CAAC;AAED,qGAAqG;AACrG,MAAM,UAAU,aAAa,CAAC,MAAsB,EAAE,KAAqB;IACzE,MAAM,EAAE,OAAO,EAAE,OAAO,EAAE,GAAG,mBAAmB,CAAC,MAAM,CAAC,eAAe,CAAC,CAAA;IACxE,MAAM,GAAG,GAAmB,EAAE,GAAG,MAAM,EAAE,CAAA;IACzC,MAAM,IAAI,GAAa,EAAE,CAAA;IACzB,MAAM,QAAQ,GAAa,EAAE,CAAA;IAC7B,MAAM,KAAK,GAAG,gBAAgB,CAAC,OAAO,CAAC,CAAA;IACvC,MAAM,UAAU,GAAG,mBAAmB,CAAC,OAAO,CAAC,CAAA;IAC/C,MAAM,EAAE,GAAG,OAAO,CAAC,UAAU,CAAC,IAAI,CAAC,CAAA;IACnC,MAAM,IAAI,GAAG,CAAC,GAAG,KAAe,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,CAAA;IAExD,0DAA0D;IAC1D,IAAI,GAAG,CAAC,eAAe,EAAE,CAAC;QACxB,GAAG,CAAC,QAAQ,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,gBAAgB,EAAE,CAAC,CAAC,CAAA;QAChD,GAAG,CAAC,aAAa,GAAG,IAAI,CAAA;QACxB,GAAG,CAAC,cAAc,GAAG,IAAI,CAAA;IAC3B,CAAC;IACD,+FAA+F;IAC/F,+FAA+F;IAC/F,kGAAkG;IAClG,IAAI,KAAK,CAAC,QAAQ,KAAK,UAAU,EAAE,CAAC;QAClC,GAAG,CAAC,GAAG,GAAG,KAAK,CAAA;QACf,GAAG,CAAC,MAAM,GAAG,KAAK,CAAA;IACpB,CAAC;IACD,IAAI,KAAK,CAAC,QAAQ,KAAK,UAAU,IAAI,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,GAAG,CAAC,UAAU,KAAK,MAAM,EAAE,CAAC;QAC7F,QAAQ,CAAC,IAAI,CAAC,mBAAmB,OAAO,2DAA2D,CAAC,CAAA;QACpG,GAAG,CAAC,UAAU,GAAG,KAAK,CAAA;IACxB,CAAC;IAED,MAAM,iBAAiB,GAAG,CAAC,KAAa,EAAU,EAAE;QAClD,IAAI,oBAAoB,CAAC,QAAQ,CAAC,KAAK,CAAC;YAAE,OAAO,KAAK,CAAA;QACtD,IAAI,KAAK,CAAC,QAAQ,KAAK,UAAU,IAAI,UAAU;YAAE,OAAO,KAAK,CAAA;QAC7D,QAAQ,CAAC,IAAI,CACX,eAAe,KAAK,gDAAgD,OAAO,IAAI,OAAO,wBAAwB,CAC/G,CAAA;QACD,OAAO,MAAM,CAAA;IACf,CAAC,CAAA;IAED,IAAI,CAAC,EAAE;QAAE,IAAI,CAAC,YAAY,CAAC,CAAA;IAC3B,IAAI,CAAC,SAAS,CAAC,CAAA;IACf,IAAI,CAAC,IAAI,EAAE,KAAK,CAAC,SAAS,CAAC,CAAA;IAE3B,IAAI,GAAG,CAAC,OAAO;QAAE,IAAI,CAAC,WAAW,CAAC,CAAA;IAClC,IAAI,GAAG,CAAC,SAAS,GAAG,CAAC;QAAE,IAAI,CAAC,aAAa,EAAE,MAAM,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,CAAA;IACjE,IAAI,GAAG,CAAC,wBAAwB;QAAE,IAAI,CAAC,mBAAmB,EAAE,GAAG,CAAC,wBAAwB,CAAC,CAAA;IAEzF,IAAI,KAAK,CAAC,UAAU,EAAE,CAAC;QACrB,IAAI,CAAC,UAAU,EAAE,KAAK,CAAC,UAAU,CAAC,CAAA;QAClC,IAAI,CAAC,GAAG,CAAC,cAAc;YAAE,IAAI,CAAC,qBAAqB,CAAC,CAAA;IACtD,CAAC;IAED,IAAI,CAAC,IAAI,EAAE,KAAK,CAAC,OAAO,CAAC,CAAA;IACzB,IAAI,CAAC,QAAQ,EAAE,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAA;IAElC,IAAI,GAAG,CAAC,aAAa;QAAE,IAAI,CAAC,iBAAiB,EAAE,GAAG,CAAC,aAAa,CAAC,CAAA;IAEjE,MAAM,GAAG,GAAG,GAAG,CAAC,YAAY,IAAI,CAAC,IAAI,GAAG,CAAC,YAAY,KAAK,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,CAAC,CAAA;IACrF,IAAI,CAAC,MAAM,EAAE,MAAM,CAAC,GAAG,CAAC,CAAC,CAAA;IAEzB,IAAI,GAAG,CAAC,OAAO,GAAG,CAAC;QAAE,IAAI,CAAC,WAAW,EAAE,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAA;IAC3D,IAAI,GAAG,CAAC,aAAa,GAAG,CAAC;QAAE,IAAI,CAAC,iBAAiB,EAAE,MAAM,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC,CAAA;IAE7E,IAAI,GAAG,CAAC,UAAU,GAAG,CAAC,IAAI,GAAG,CAAC,UAAU,KAAK,IAAI;QAAE,IAAI,CAAC,cAAc,EAAE,MAAM,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAA;IAC/F,IAAI,GAAG,CAAC,WAAW,GAAG,CAAC,IAAI,GAAG,CAAC,WAAW,KAAK,GAAG;QAAE,IAAI,CAAC,eAAe,EAAE,MAAM,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,CAAA;IAElG,IAAI,GAAG,CAAC,MAAM,KAAK,EAAE;QAAE,IAAI,CAAC,UAAU,EAAE,GAAG,CAAC,MAAM,CAAC,CAAA;IACnD,IAAI,GAAG,CAAC,UAAU,KAAK,EAAE,IAAI,GAAG,CAAC,UAAU,KAAK,OAAO;QAAE,IAAI,CAAC,cAAc,EAAE,GAAG,CAAC,UAAU,CAAC,CAAA;IAC7F,IAAI,GAAG,CAAC,QAAQ,KAAK,CAAC;QAAE,IAAI,CAAC,YAAY,EAAE,MAAM,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAA;IAEhE,mBAAmB;IACnB,IAAI,EAAE,EAAE,CAAC;QACP,IAAI,GAAG,CAAC,UAAU,KAAK,IAAI;YAAE,IAAI,CAAC,KAAK,CAAC,CAAA;IAC1C,CAAC;SAAM,CAAC;QACN,MAAM,SAAS,GAAG,UAAU,IAAI,CAAC,KAAK,KAAK,SAAS,IAAI,KAAK,IAAI,+BAA+B,CAAC,CAAA;QACjG,IAAI,SAAS,EAAE,CAAC;YACd,IAAI,GAAG,CAAC,UAAU,KAAK,MAAM,IAAI,GAAG,CAAC,UAAU,KAAK,IAAI,IAAI,GAAG,CAAC,UAAU,KAAK,KAAK,EAAE,CAAC;gBACrF,IAAI,CAAC,cAAc,EAAE,GAAG,CAAC,UAAU,CAAC,CAAA;YACtC,CAAC;QACH,CAAC;aAAM,IAAI,GAAG,CAAC,UAAU,KAAK,IAAI,EAAE,CAAC;YACnC,IAAI,CAAC,cAAc,CAAC,CAAA;QACtB,CAAC;IACH,CAAC;IAED,IAAI,GAAG,CAAC,SAAS;QAAE,IAAI,CAAC,iBAAiB,CAAC,CAAA;IAC1C,IAAI,GAAG,CAAC,aAAa;QAAE,IAAI,CAAC,iBAAiB,CAAC,CAAA;IAC9C,IAAI,GAAG,CAAC,OAAO;QAAE,IAAI,CAAC,WAAW,CAAC,CAAA;IAClC,IAAI,GAAG,CAAC,KAAK;QAAE,IAAI,CAAC,SAAS,CAAC,CAAA;IAC9B,IAAI,GAAG,CAAC,aAAa;QAAE,IAAI,CAAC,iBAAiB,CAAC,CAAA;IAE9C,IAAI,GAAG,CAAC,QAAQ,GAAG,CAAC,EAAE,CAAC;QACrB,IAAI,CAAC,YAAY,EAAE,MAAM,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAA;QACxC,IAAI,GAAG,CAAC,QAAQ,KAAK,CAAC;YAAE,IAAI,CAAC,MAAM,CAAC,CAAA,CAAC,2BAA2B;IAClE,CAAC;IAED,uDAAuD;IACvD,MAAM,MAAM,GAAG,GAAG,EAAE;QAClB,IAAI,CAAC,GAAG,CAAC,GAAG,IAAI,KAAK,CAAC,WAAW;YAAE,OAAM;QACzC,IAAI,GAAG,CAAC,cAAc,KAAK,EAAE,EAAE,CAAC;YAC9B,IAAI,CAAC,CAAC,KAAK,KAAK,SAAS,IAAI,KAAK,IAAI,mBAAmB,CAAC,EAAE,CAAC;gBAC3D,QAAQ,CAAC,IAAI,CACX,2CAA2C,OAAO,IAAI,OAAO,mCAAmC,mBAAmB,aAAa,CACjI,CAAA;gBACD,OAAM;YACR,CAAC;YACD,MAAM,CAAC,GAAG,GAAG,CAAC,YAAY,CAAA;YAC1B,MAAM,CAAC,GAAG,GAAG,CAAC,YAAY,CAAA;YAC1B,IAAI,CAAC,CAAC,KAAK,EAAE,IAAI,CAAC,KAAK,KAAK,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,IAAI,CAAC,KAAK,KAAK,CAAC,EAAE,CAAC;gBAC3D,QAAQ,CAAC,IAAI,CACX,qDAAqD,CAAC,IAAI,KAAK,OAAO,CAAC,IAAI,KAAK,8BAA8B,CAC/G,CAAA;YACH,CAAC;YACD,IAAI,CAAC,eAAe,EAAE,GAAG,CAAC,cAAc,EAAE,aAAa,EAAE,WAAW,EAAE,oBAAoB,EAAE,GAAG,CAAC,CAAA;YAChG,OAAM;QACR,CAAC;QACD,IAAI,CAAC,CAAC,KAAK,KAAK,SAAS,IAAI,KAAK,IAAI,aAAa,CAAC,EAAE,CAAC;YACrD,QAAQ,CAAC,IAAI,CACX,mCAAmC,OAAO,IAAI,OAAO,kCAAkC,aAAa,aAAa,CAClH,CAAA;YACD,OAAM;QACR,CAAC;QACD,IAAI,CAAC,aAAa,EAAE,WAAW,EAAE,oBAAoB,EAAE,GAAG,CAAC,CAAA;IAC7D,CAAC,CAAA;IACD,MAAM,SAAS,GAAG,GAAG,EAAE;QACrB,IAAI,CAAC,GAAG,CAAC,MAAM,IAAI,KAAK,CAAC,WAAW;YAAE,OAAM;QAC5C,IAAI,GAAG,CAAC,iBAAiB,KAAK,EAAE,EAAE,CAAC;YACjC,QAAQ,CAAC,IAAI,CAAC,0DAA0D,CAAC,CAAA;YACzE,OAAM;QACR,CAAC;QACD,IAAI,CAAC,GAAG,CAAC,qBAAqB,EAAE,CAAC;YAC/B,QAAQ,CAAC,IAAI,CACX,gCAAgC,OAAO,IAAI,OAAO,+FAA+F,CAClJ,CAAA;YACD,OAAM;QACR,CAAC;QACD,IAAI,CACF,eAAe,EACf,GAAG,CAAC,iBAAiB,EACrB,aAAa,EACb,cAAc,EACd,oBAAoB,EACpB,MAAM,CAAC,GAAG,CAAC,YAAY,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,CAAC,CACrD,CAAA;IACH,CAAC,CAAA;IACD,IAAI,GAAG,CAAC,MAAM,IAAI,GAAG,CAAC,GAAG,EAAE,CAAC;QAC1B,QAAQ,CAAC,IAAI,CAAC,uDAAuD,CAAC,CAAA;QACtE,SAAS,EAAE,CAAA;IACb,CAAC;SAAM,CAAC;QACN,MAAM,EAAE,CAAA;QACR,SAAS,EAAE,CAAA;IACb,CAAC;IAED,IAAI,GAAG,CAAC,cAAc;QAAE,IAAI,CAAC,WAAW,CAAC,CAAA;IAEzC,0BAA0B;IAC1B,IAAI,CAAC,KAAK,CAAC,WAAW,EAAE,CAAC;QACvB,IAAI,CAAC,GAAG,CAAC,kBAAkB,EAAE,CAAC;YAC5B,IAAI,KAAK,KAAK,SAAS,IAAI,KAAK,IAAI,uCAAuC;gBACzE,IAAI,CAAC,yBAAyB,CAAC,CAAA;QACnC,CAAC;aAAM,IAAI,KAAK,KAAK,SAAS,IAAI,KAAK,IAAI,4BAA4B,EAAE,CAAC;YACxE,IAAI,CAAC,sBAAsB,CAAC,CAAA;QAC9B,CAAC;aAAM,CAAC;YACN,QAAQ,CAAC,IAAI,CACX,sDAAsD,OAAO,IAAI,OAAO,gCAAgC,4BAA4B,aAAa,CAClJ,CAAA;QACH,CAAC;IACH,CAAC;IAED,IAAI,KAAK,CAAC,WAAW,EAAE,CAAC;QACtB,IAAI,CAAC,aAAa,EAAE,WAAW,EAAE,MAAM,CAAC,CAAA;IAC1C,CAAC;SAAM,CAAC;QACN,IAAI,GAAG,CAAC,QAAQ,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG;YAAE,IAAI,CAAC,YAAY,EAAE,MAAM,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAA;QAC1E,IAAI,GAAG,CAAC,SAAS,GAAG,CAAC;YAAE,IAAI,CAAC,aAAa,EAAE,MAAM,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,CAAA;QACjE,IAAI,GAAG,CAAC,YAAY,KAAK,EAAE,IAAI,GAAG,CAAC,YAAY,KAAK,KAAK,EAAE,CAAC;YAC1D,MAAM,KAAK,GAAG,iBAAiB,CAAC,GAAG,CAAC,YAAY,CAAC,CAAA;YACjD,IAAI,KAAK,KAAK,KAAK;gBAAE,IAAI,CAAC,gBAAgB,EAAE,KAAK,CAAC,CAAA;QACpD,CAAC;QACD,IACE,GAAG,CAAC,UAAU,KAAK,KAAK;YACxB,GAAG,CAAC,YAAY,KAAK,EAAE;YACvB,GAAG,CAAC,YAAY,KAAK,KAAK;YAC1B,GAAG,CAAC,YAAY,KAAK,KAAK,EAC1B,CAAC;YACD,MAAM,KAAK,GAAG,iBAAiB,CAAC,GAAG,CAAC,YAAY,CAAC,CAAA;YACjD,IAAI,KAAK,KAAK,KAAK,IAAI,KAAK,KAAK,KAAK;gBAAE,IAAI,CAAC,gBAAgB,EAAE,KAAK,CAAC,CAAA;QACvE,CAAC;QACD,IAAI,UAAU,CAAC,GAAG,CAAC,YAAY,EAAE,GAAG,CAAC;YAAE,IAAI,CAAC,gBAAgB,EAAE,aAAa,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC,CAAA;QAC9F,IAAI,GAAG,CAAC,YAAY,KAAK,EAAE,IAAI,GAAG,CAAC,YAAY,KAAK,MAAM;YAAE,IAAI,CAAC,gBAAgB,EAAE,GAAG,CAAC,YAAY,CAAC,CAAA;QACpG,IAAI,UAAU,CAAC,GAAG,CAAC,UAAU,EAAE,GAAG,CAAC;YAAE,IAAI,CAAC,cAAc,EAAE,aAAa,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAA;QACxF,IAAI,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,cAAc,CAAC,KAAK,CAAC;YAAE,IAAI,CAAC,kBAAkB,EAAE,aAAa,CAAC,GAAG,CAAC,cAAc,CAAC,CAAC,CAAA;QACtG,IAAI,UAAU,CAAC,GAAG,CAAC,eAAe,EAAE,GAAG,CAAC;YAAE,IAAI,CAAC,mBAAmB,EAAE,aAAa,CAAC,GAAG,CAAC,eAAe,CAAC,CAAC,CAAA;IACzG,CAAC;IAED,IAAI,CAAC,EAAE,EAAE,CAAC;QACR,IAAI,CAAC,OAAO,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,CAAA;QACrC,IAAI,GAAG,CAAC,GAAG,EAAE,CAAC;YACZ,IAAI,GAAG,CAAC,OAAO,KAAK,EAAE,IAAI,GAAG,CAAC,OAAO,KAAK,MAAM;gBAAE,IAAI,CAAC,WAAW,EAAE,GAAG,CAAC,OAAO,CAAC,CAAA;YAChF,IAAI,GAAG,CAAC,UAAU,KAAK,EAAE,IAAI,GAAG,CAAC,UAAU,KAAK,MAAM;gBAAE,IAAI,CAAC,cAAc,EAAE,GAAG,CAAC,UAAU,CAAC,CAAA;QAC9F,CAAC;IACH,CAAC;IAED,IAAI,CAAC;QACH,IAAI,CAAC,GAAG,cAAc,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAA;IACzC,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,QAAQ,CAAC,IAAI,CAAC,kDAAmD,KAAe,CAAC,OAAO,EAAE,CAAC,CAAA;IAC7F,CAAC;IAED,gGAAgG;IAChG,+EAA+E;IAC/E,MAAM,GAAG,GAAG,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,CAAC,CAAA;IACxD,IAAI,GAAG,IAAI,CAAC,EAAE,CAAC;QACb,MAAM,IAAI,eAAe,CACvB,kBAAkB,EAClB,0CAA0C,EAC1C,YAAY,GAAG,WAAW,IAAI,CAAC,GAAG,GAAG,CAAC,CAAC,IAAI,EAAE,QAAQ,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CACzE,CAAA;IACH,CAAC;IAED,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE,CAAA;AAC7C,CAAC;AAED,8BAA8B;AAC9B,MAAM,UAAU,cAAc,CAAC,MAAsB,EAAE,KAAqB;IAC1E,OAAO,aAAa,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC,IAAI,CAAA;AAC1C,CAAC"}

package/dist/runtime/llamacpp/ctx-ladder.d.ts

@@ -0,0 +1,16 @@
+/**
+ * The context-window ladder, ported from the app's `computeNextCtxLen`
+ * (`core/src/browser/models/utils.ts`).
+ *
+ * When a request overflows the model's context the app does not fail it — it reloads the model with
+ * a bigger window and retries. The steps are `<8192 → 8192 → 32768 → ×1.5`, capped at what the model
+ * was trained for. They are not derived from anything; they are the steps the UI has always used,
+ * and the ladder is shared so a session that grew under the app and one that grew under the core end
+ * up at the same sizes.
+ */
+/**
+ * The context window assumed when nothing else says otherwise — the same single value the app
+ * settled on after five different defaults disagreed across its tree.
+ */
+export declare const DEFAULT_CTX_LEN = 16384;
+export declare function computeNextCtxLen(currentCtxLen: number, maxCtxLen?: number): number;

package/dist/runtime/llamacpp/ctx-ladder.js

@@ -0,0 +1,28 @@
+/**
+ * The context-window ladder, ported from the app's `computeNextCtxLen`
+ * (`core/src/browser/models/utils.ts`).
+ *
+ * When a request overflows the model's context the app does not fail it — it reloads the model with
+ * a bigger window and retries. The steps are `<8192 → 8192 → 32768 → ×1.5`, capped at what the model
+ * was trained for. They are not derived from anything; they are the steps the UI has always used,
+ * and the ladder is shared so a session that grew under the app and one that grew under the core end
+ * up at the same sizes.
+ */
+/**
+ * The context window assumed when nothing else says otherwise — the same single value the app
+ * settled on after five different defaults disagreed across its tree.
+ */
+export const DEFAULT_CTX_LEN = 16384;
+export function computeNextCtxLen(currentCtxLen, maxCtxLen) {
+    let next;
+    if (currentCtxLen < 8192)
+        next = 8192;
+    else if (currentCtxLen < 32768)
+        next = 32768;
+    else
+        next = Math.round(currentCtxLen * 1.5);
+    if (typeof maxCtxLen === 'number' && maxCtxLen > 0)
+        next = Math.min(next, maxCtxLen);
+    return next;
+}
+//# sourceMappingURL=ctx-ladder.js.map

package/dist/runtime/llamacpp/ctx-ladder.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"ctx-ladder.js","sourceRoot":"","sources":["../../../src/runtime/llamacpp/ctx-ladder.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH;;;GAGG;AACH,MAAM,CAAC,MAAM,eAAe,GAAG,KAAK,CAAA;AAEpC,MAAM,UAAU,iBAAiB,CAAC,aAAqB,EAAE,SAAkB;IACzE,IAAI,IAAY,CAAA;IAChB,IAAI,aAAa,GAAG,IAAI;QAAE,IAAI,GAAG,IAAI,CAAA;SAChC,IAAI,aAAa,GAAG,KAAK;QAAE,IAAI,GAAG,KAAK,CAAA;;QACvC,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,aAAa,GAAG,GAAG,CAAC,CAAA;IAC3C,IAAI,OAAO,SAAS,KAAK,QAAQ,IAAI,SAAS,GAAG,CAAC;QAAE,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,SAAS,CAAC,CAAA;IACpF,OAAO,IAAI,CAAA;AACb,CAAC"}

package/dist/runtime/llamacpp/devices.d.ts

@@ -0,0 +1,26 @@
+/**
+ * Parser for `llama-server --list-devices` stdout. Port of the parsing half of
+ * `src-tauri/plugins/tauri-plugin-llamacpp-upstream/src/device.rs`; spawning lives in `runtime/`.
+ * Pinned by `test/contract/devices.test.ts`.
+ *
+ * Accepted line shapes (everything after the `Available devices:` header, blank lines skipped):
+ *   `Vulkan0: Intel(R) Arc(tm) A750 Graphics (DG2) (8128 MiB, 8128 MiB free)`
+ *   `CUDA0: NVIDIA GeForce RTX 4090 (24576 MiB, 24000 MiB free)`
+ * The last parenthesised group that looks like `<n> MiB, <n> MiB free` is the memory; the name is
+ * everything before it. Unparseable lines are skipped, a missing header is an error, and an empty
+ * list is a valid result (the caller treats it as "this tier cannot enumerate a GPU").
+ */
+import type { DeviceInfo } from '../../contracts/index.js';
+export declare const DEVICES_HEADER = "Available devices:";
+/** `"8128 MiB, 8128 MiB free"` — both halves start with an i32 and mention `MiB`. */
+export declare function isMemoryPattern(content: string): boolean;
+/** Index of the `(` opening the last memory-looking group, and its content. */
+export declare function findMemoryPattern(text: string): {
+    start: number;
+    content: string;
+} | undefined;
+/** First whitespace token as i32 (`"7721 MiB free"` → 7721). */
+export declare function parseMemoryValue(memStr: string): number | undefined;
+export declare function parseDeviceLine(rawLine: string): DeviceInfo | undefined;
+/** Throws `DEVICE_LIST_PARSE_FAILED` (details = the whole output) when the header is missing. */
+export declare function parseDeviceOutput(output: string): DeviceInfo[];

package/dist/runtime/llamacpp/devices.js

@@ -0,0 +1,86 @@
+/**
+ * Parser for `llama-server --list-devices` stdout. Port of the parsing half of
+ * `src-tauri/plugins/tauri-plugin-llamacpp-upstream/src/device.rs`; spawning lives in `runtime/`.
+ * Pinned by `test/contract/devices.test.ts`.
+ *
+ * Accepted line shapes (everything after the `Available devices:` header, blank lines skipped):
+ *   `Vulkan0: Intel(R) Arc(tm) A750 Graphics (DG2) (8128 MiB, 8128 MiB free)`
+ *   `CUDA0: NVIDIA GeForce RTX 4090 (24576 MiB, 24000 MiB free)`
+ * The last parenthesised group that looks like `<n> MiB, <n> MiB free` is the memory; the name is
+ * everything before it. Unparseable lines are skipped, a missing header is an error, and an empty
+ * list is a valid result (the caller treats it as "this tier cannot enumerate a GPU").
+ */
+import { AtomicCoreError } from '../../contracts/index.js';
+import { parseRustI32 } from '../../util/index.js';
+export const DEVICES_HEADER = 'Available devices:';
+/** `"8128 MiB, 8128 MiB free"` — both halves start with an i32 and mention `MiB`. */
+export function isMemoryPattern(content) {
+    if (!(content.includes('MiB') && content.includes('free') && content.includes(',')))
+        return false;
+    const parts = content.split(',');
+    if (parts.length !== 2)
+        return false;
+    return parts.every((raw) => {
+        const part = raw.trim();
+        const first = part.split(/\s+/).find((t) => t !== '');
+        return first !== undefined && parseRustI32(first) !== undefined && part.includes('MiB');
+    });
+}
+/** Index of the `(` opening the last memory-looking group, and its content. */
+export function findMemoryPattern(text) {
+    let last;
+    for (let i = 0; i < text.length; i++) {
+        if (text[i] !== '(')
+            continue;
+        const remaining = text.slice(i + 1);
+        const close = remaining.indexOf(')');
+        if (close < 0)
+            continue;
+        const content = remaining.slice(0, close);
+        if (isMemoryPattern(content))
+            last = { start: i, content };
+    }
+    return last;
+}
+/** First whitespace token as i32 (`"7721 MiB free"` → 7721). */
+export function parseMemoryValue(memStr) {
+    const first = memStr.split(/\s+/).find((t) => t !== '');
+    return first === undefined ? undefined : parseRustI32(first);
+}
+export function parseDeviceLine(rawLine) {
+    const line = rawLine.trim();
+    const colon = line.indexOf(':');
+    if (colon < 0)
+        return undefined;
+    const id = line.slice(0, colon).trim();
+    const rest = line.slice(colon + 1).trim();
+    const memory = findMemoryPattern(rest);
+    if (memory === undefined)
+        return undefined;
+    const name = rest.slice(0, memory.start).trim();
+    const [total, free] = memory.content.split(',').map((p) => parseMemoryValue(p.trim()));
+    if (total === undefined || free === undefined)
+        return undefined;
+    return { id, name, mem: total, free };
+}
+/** Throws `DEVICE_LIST_PARSE_FAILED` (details = the whole output) when the header is missing. */
+export function parseDeviceOutput(output) {
+    const devices = [];
+    let foundSection = false;
+    for (const raw of output.split(/\r?\n/)) {
+        if (raw.trim() === DEVICES_HEADER) {
+            foundSection = true;
+            continue;
+        }
+        if (!foundSection || raw.trim() === '')
+            continue;
+        const device = parseDeviceLine(raw);
+        if (device !== undefined)
+            devices.push(device);
+    }
+    if (!foundSection) {
+        throw new AtomicCoreError('DEVICE_LIST_PARSE_FAILED', "Could not find 'Available devices:' section in the backend output.", output);
+    }
+    return devices;
+}
+//# sourceMappingURL=devices.js.map

package/dist/runtime/llamacpp/devices.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"devices.js","sourceRoot":"","sources":["../../../src/runtime/llamacpp/devices.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,EAAE,eAAe,EAAE,MAAM,0BAA0B,CAAA;AAE1D,OAAO,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAA;AAElD,MAAM,CAAC,MAAM,cAAc,GAAG,oBAAoB,CAAA;AAElD,qFAAqF;AACrF,MAAM,UAAU,eAAe,CAAC,OAAe;IAC7C,IAAI,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;QAAE,OAAO,KAAK,CAAA;IACjG,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,CAAA;IAChC,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,KAAK,CAAA;IACpC,OAAO,KAAK,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,EAAE;QACzB,MAAM,IAAI,GAAG,GAAG,CAAC,IAAI,EAAE,CAAA;QACvB,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,KAAK,EAAE,CAAC,CAAA;QACrD,OAAO,KAAK,KAAK,SAAS,IAAI,YAAY,CAAC,KAAK,CAAC,KAAK,SAAS,IAAI,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAA;IACzF,CAAC,CAAC,CAAA;AACJ,CAAC;AAED,+EAA+E;AAC/E,MAAM,UAAU,iBAAiB,CAAC,IAAY;IAC5C,IAAI,IAAoD,CAAA;IACxD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACrC,IAAI,IAAI,CAAC,CAAC,CAAC,KAAK,GAAG;YAAE,SAAQ;QAC7B,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,CAAA;QACnC,MAAM,KAAK,GAAG,SAAS,CAAC,OAAO,CAAC,GAAG,CAAC,CAAA;QACpC,IAAI,KAAK,GAAG,CAAC;YAAE,SAAQ;QACvB,MAAM,OAAO,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,CAAA;QACzC,IAAI,eAAe,CAAC,OAAO,CAAC;YAAE,IAAI,GAAG,EAAE,KAAK,EAAE,CAAC,EAAE,OAAO,EAAE,CAAA;IAC5D,CAAC;IACD,OAAO,IAAI,CAAA;AACb,CAAC;AAED,gEAAgE;AAChE,MAAM,UAAU,gBAAgB,CAAC,MAAc;IAC7C,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,KAAK,EAAE,CAAC,CAAA;IACvD,OAAO,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,YAAY,CAAC,KAAK,CAAC,CAAA;AAC9D,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,OAAe;IAC7C,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,EAAE,CAAA;IAC3B,MAAM,KAAK,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAA;IAC/B,IAAI,KAAK,GAAG,CAAC;QAAE,OAAO,SAAS,CAAA;IAC/B,MAAM,EAAE,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,IAAI,EAAE,CAAA;IACtC,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,CAAA;IACzC,MAAM,MAAM,GAAG,iBAAiB,CAAC,IAAI,CAAC,CAAA;IACtC,IAAI,MAAM,KAAK,SAAS;QAAE,OAAO,SAAS,CAAA;IAC1C,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,EAAE,CAAA;IAC/C,MAAM,CAAC,KAAK,EAAE,IAAI,CAAC,GAAG,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,gBAAgB,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAA;IACtF,IAAI,KAAK,KAAK,SAAS,IAAI,IAAI,KAAK,SAAS;QAAE,OAAO,SAAS,CAAA;IAC/D,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,GAAG,EAAE,KAAK,EAAE,IAAI,EAAE,CAAA;AACvC,CAAC;AAED,iGAAiG;AACjG,MAAM,UAAU,iBAAiB,CAAC,MAAc;IAC9C,MAAM,OAAO,GAAiB,EAAE,CAAA;IAChC,IAAI,YAAY,GAAG,KAAK,CAAA;IACxB,KAAK,MAAM,GAAG,IAAI,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,CAAC;QACxC,IAAI,GAAG,CAAC,IAAI,EAAE,KAAK,cAAc,EAAE,CAAC;YAClC,YAAY,GAAG,IAAI,CAAA;YACnB,SAAQ;QACV,CAAC;QACD,IAAI,CAAC,YAAY,IAAI,GAAG,CAAC,IAAI,EAAE,KAAK,EAAE;YAAE,SAAQ;QAChD,MAAM,MAAM,GAAG,eAAe,CAAC,GAAG,CAAC,CAAA;QACnC,IAAI,MAAM,KAAK,SAAS;YAAE,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAA;IAChD,CAAC;IACD,IAAI,CAAC,YAAY,EAAE,CAAC;QAClB,MAAM,IAAI,eAAe,CACvB,0BAA0B,EAC1B,oEAAoE,EACpE,MAAM,CACP,CAAA;IACH,CAAC;IACD,OAAO,OAAO,CAAA;AAChB,CAAC"}

package/dist/runtime/llamacpp/errors.d.ts

@@ -0,0 +1,34 @@
+/**
+ * Classification of a failed `llama-server` process into the app's error contract. Port of
+ * `src-tauri/plugins/tauri-plugin-llamacpp-upstream/src/error.rs` (`LlamacppError::from_stderr`,
+ * `from_exit_status`, `from_process_output`). Pinned by `test/contract/errors.test.ts`. The
+ * TurboQuant fork's `error.rs` differs in one rule, `wrong number of tensors` (app commit
+ * `ec1fd3ea7`), so the provider is a parameter.
+ *
+ * The cascade is ordered: the first matching group wins. Messages are the app's user-facing strings
+ * and are part of the contract. `details` is always set (the raw stream), even when empty.
+ */
+import { AtomicCoreError } from '../../contracts/index.js';
+import type { LocalProviderId } from '../../contracts/index.js';
+export interface ExitInfo {
+    /** Exit code, or null when the process was killed by a signal. Windows codes may be unsigned. */
+    code: number | null;
+    /** Unix signal as number (11) or Node name ('SIGSEGV'); null on a normal exit. */
+    signal: number | string | null;
+}
+/** Normalise a Node signal name to the Unix number Rust sees. */
+export declare function signalNumber(signal: number | string | null): number | null;
+/** A hard native crash (segfault / abort) rather than a normal non-zero exit. */
+export declare function isCrashExit(exit: ExitInfo, platform: NodeJS.Platform): boolean;
+export declare const GENERIC_PROCESS_ERROR_MESSAGE = "The model process encountered an unexpected error.";
+export declare const CRASH_MESSAGE = "The model process crashed unexpectedly (access violation / segfault). This usually means the model is incompatible with this backend, or its speculative-decoding (MTP) configuration is unsupported here.";
+/** `LlamacppError::from_stderr`: ordered substring cascade over the lowercased stream. */
+export declare function classifyStderr(stderr: string, provider?: LocalProviderId): AtomicCoreError;
+/** `LlamacppError::from_exit_status`: a recognised crash upgrades the generic error's message. */
+export declare function classifyExit(exit: ExitInfo, stderr: string, platform: NodeJS.Platform, provider?: LocalProviderId): AtomicCoreError;
+/**
+ * `LlamacppError::from_process_output`: stderr classification wins; otherwise try stdout (several
+ * llama.cpp builds report loader failures there); otherwise keep the exit-status error but hand the
+ * caller stdout as details when stderr is blank.
+ */
+export declare function classifyProcessOutput(exit: ExitInfo, stderr: string, stdout: string, platform: NodeJS.Platform, provider?: LocalProviderId): AtomicCoreError;