@seanhogg/builderforce-memory-engine 2026.6.27 → 2026.6.28
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/dist/index.d.ts +6 -0
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +5 -0
- package/dist/index.js.map +1 -1
- package/dist/lm/evermind_lm.d.ts +148 -0
- package/dist/lm/evermind_lm.d.ts.map +1 -0
- package/dist/lm/evermind_lm.js +479 -0
- package/dist/lm/evermind_lm.js.map +1 -0
- package/dist/lm/index.d.ts +6 -0
- package/dist/lm/index.d.ts.map +1 -0
- package/dist/lm/index.js +5 -0
- package/dist/lm/index.js.map +1 -0
- package/dist/model/attention_block.js +1 -1
- package/dist/model/attention_block.js.map +1 -1
- package/dist/model/mamba_model.js +1 -1
- package/dist/model/mamba_model.js.map +1 -1
- package/dist/moe/index.d.ts +10 -0
- package/dist/moe/index.d.ts.map +1 -0
- package/dist/moe/index.js +7 -0
- package/dist/moe/index.js.map +1 -0
- package/dist/moe/moe_model.d.ts +134 -0
- package/dist/moe/moe_model.d.ts.map +1 -0
- package/dist/moe/moe_model.js +415 -0
- package/dist/moe/moe_model.js.map +1 -0
- package/dist/moe/moe_package.d.ts +81 -0
- package/dist/moe/moe_package.d.ts.map +1 -0
- package/dist/moe/moe_package.js +157 -0
- package/dist/moe/moe_package.js.map +1 -0
- package/dist/moe/moe_trainer.d.ts +53 -0
- package/dist/moe/moe_trainer.d.ts.map +1 -0
- package/dist/moe/moe_trainer.js +93 -0
- package/dist/moe/moe_trainer.js.map +1 -0
- package/dist/optim/adamw.d.ts +32 -0
- package/dist/optim/adamw.d.ts.map +1 -0
- package/dist/optim/adamw.js +52 -0
- package/dist/optim/adamw.js.map +1 -0
- package/package.json +1 -1
- package/src/index.ts +28 -0
- package/src/lm/evermind_lm.ts +558 -0
- package/src/lm/index.ts +6 -0
- package/src/model/attention_block.ts +1 -1
- package/src/model/mamba_model.ts +1 -1
- package/src/moe/index.ts +23 -0
- package/src/moe/moe_model.ts +475 -0
- package/src/moe/moe_package.ts +205 -0
- package/src/moe/moe_trainer.ts +134 -0
- package/src/optim/adamw.ts +72 -0
package/dist/index.d.ts
CHANGED
|
@@ -7,6 +7,12 @@ export { Mamba2Block } from './model/mamba2_block.js';
|
|
|
7
7
|
export { Mamba3Block } from './model/mamba3_block.js';
|
|
8
8
|
export { AttentionBlock } from './model/attention_block.js';
|
|
9
9
|
export { MambaBlock } from './model/mamba1_block.js';
|
|
10
|
+
export { SharedExpertMoE, LoadBalanceAccumulator, DEFAULT_MOE_CONFIG, DEFAULT_MOE_SEED, MoETrainer, EvermindModelPackage, } from './moe/index.js';
|
|
11
|
+
export type { MoEConfig, MoEParam, RouteResult, MoESample, MoETrainOptions, MoEEpochResult, EvermindModelManifest, EvermindModelCard, PackageMeta, ValidationResult, } from './moe/index.js';
|
|
12
|
+
export { EvermindLM, EvermindLMTrainer, DEFAULT_LM_CONFIG, DEFAULT_LM_SEED } from './lm/index.js';
|
|
13
|
+
export type { EvermindLMConfig, LMGenerateOptions, TextCodec } from './lm/index.js';
|
|
14
|
+
export { AdamW } from './optim/adamw.js';
|
|
15
|
+
export type { AdamWOptions, OptimTarget, OptimParam } from './optim/adamw.js';
|
|
10
16
|
export { MambaTrainer } from './training/trainer.js';
|
|
11
17
|
export { Tensor, backward, enableGrad, noGrad, clearTape, recordOperation, crossEntropyLoss, crossEntropyGrad, } from './training/autograd.js';
|
|
12
18
|
export { BPETokenizer } from './tokenizer/bpe.js';
|
package/dist/index.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAIH,OAAO,EAAE,gBAAgB,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAGtE,OAAO,EAAE,WAAW,EAAE,MAAQ,yBAAyB,CAAC;AACxD,OAAO,EAAE,WAAW,EAAE,MAAQ,yBAAyB,CAAC;AACxD,OAAO,EAAE,WAAW,EAAE,MAAQ,yBAAyB,CAAC;AACxD,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAG5D,OAAO,EAAE,UAAU,EAAE,MAAM,yBAAyB,CAAC;
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAIH,OAAO,EAAE,gBAAgB,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAGtE,OAAO,EAAE,WAAW,EAAE,MAAQ,yBAAyB,CAAC;AACxD,OAAO,EAAE,WAAW,EAAE,MAAQ,yBAAyB,CAAC;AACxD,OAAO,EAAE,WAAW,EAAE,MAAQ,yBAAyB,CAAC;AACxD,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAG5D,OAAO,EAAE,UAAU,EAAE,MAAM,yBAAyB,CAAC;AAGrD,OAAO,EACH,eAAe,EACf,sBAAsB,EACtB,kBAAkB,EAClB,gBAAgB,EAChB,UAAU,EACV,oBAAoB,GACvB,MAAM,gBAAgB,CAAC;AACxB,YAAY,EACR,SAAS,EACT,QAAQ,EACR,WAAW,EACX,SAAS,EACT,eAAe,EACf,cAAc,EACd,qBAAqB,EACrB,iBAAiB,EACjB,WAAW,EACX,gBAAgB,GACnB,MAAM,gBAAgB,CAAC;AAGxB,OAAO,EAAE,UAAU,EAAE,iBAAiB,EAAE,iBAAiB,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAClG,YAAY,EAAE,gBAAgB,EAAE,iBAAiB,EAAE,SAAS,EAAE,MAAM,eAAe,CAAC;AACpF,OAAO,EAAE,KAAK,EAAE,MAAM,kBAAkB,CAAC;AACzC,YAAY,EAAE,YAAY,EAAE,WAAW,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAI9E,OAAO,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAC;AACrD,OAAO,EACH,MAAM,EACN,QAAQ,EACR,UAAU,EACV,MAAM,EACN,SAAS,EACT,eAAe,EACf,gBAAgB,EAChB,gBAAgB,GACnB,MAAM,wBAAwB,CAAC;AAIhC,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAIlD,OAAO,EAAE,SAAS,EAAE,WAAW,EAAE,KAAK,EAAE,aAAa,EAAE,MAAM,gBAAgB,CAAC;AAI9E,YAAY,EACR,sBAAsB,EACtB,gBAAgB,EAChB,kBAAkB,EAClB,eAAe,EACf,SAAS,GACZ,MAAM,wBAAwB,CAAC;AAEhC,YAAY,EAAE,aAAa,EAAE,UAAU,EAAE,SAAS,EAAE,kBAAkB,EAAE,MAAM,2BAA2B,CAAC;AAC1G,YAAY,EAAE,iBAAiB,EAAE,UAAU,EAAE,UAAU,EAAE,kBAAkB,EAAE,gBAAgB,EAAE,MAAM,yBAAyB,CAAC;AAC/H,YAAY,EAAE,iBAAiB,EAAE,WAAW,EAAE,MAAM,yBAAyB,CAAC;AAC9E,YAAY,EAAE,iBAAiB,EAAE,WAAW,EAAE,MAAO,yBAAyB,CAAC;AAC/E,YAAY,EAAE,oBAAoB,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAIvF,OAAO,EACH,UAAU,EACV,mBAAmB,EACnB,wBAAwB,EACxB,mBAAmB,EACnB,qBAAqB,EACrB,eAAe,EACf,cAAc,EACd,UAAU,EACV,YAAY,EACZ,IAAI,GACP,MAAM,sBAAsB,CAAC;AAI9B,OAAO,EACH,YAAY,EACZ,cAAc,EACd,WAAW,EACX,WAAW,EACX,YAAY,EACZ,cAAc,EACd,sBAAsB,EACtB,wBAAwB,EACxB,cAAc,GACjB,MAAM,yBAAyB,CAAC;AAKjC,OAAO,EAAE,2BAA2B,EAAE,4BAA4B,EAAE,MAC3D,6BAA6B,CAAC;AACvC,OAAO,EAAE,mBAAmB,EAAE,oBAAoB,EAAE,MAC3C,qBAAqB,CAAC;AAC/B,OAAO,EAAE,mBAAmB,EAAE,oBAAoB,EAAE,MAC3C,gCAAgC,CAAC;AAC1C,OAAO,EAAE,kBAAkB,EAAE,cAAc,EAAE,MACpC,4BAA4B,CAAC;AACtC,OAAO,EAAE,gBAAgB,EAAE,yBAAyB,EAAE,oBAAoB,EAAE,qBAAqB,EAAE,MAC1F,0BAA0B,CAAC;AAGpC,OAAO,EAAE,gBAAgB,EAAE,iBAAiB,EAAE,MACrC,kBAAkB,CAAC;AAG5B,OAAO,EAAE,wBAAwB,EAAE,yBAAyB,EAAE,MACrD,0BAA0B,CAAC;AAGpC,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,YAAY,EAAE,MAC/D,wBAAwB,CAAC;AAIlC,OAAO,EACH,MAAM,EACN,UAAU,EACV,gBAAgB,EAChB,gBAAgB,EAChB,aAAa,EACb,aAAa,EACb,QAAQ,EACR,UAAU,EACV,YAAY,EACZ,aAAa,EACb,aAAa,EACb,WAAW,EACX,qBAAqB,EACrB,mBAAmB,EACnB,aAAa,GAChB,MAAM,mBAAmB,CAAC;AAC3B,YAAY,EACR,MAAM,EACN,aAAa,EACb,iBAAiB,EACjB,aAAa,EACb,WAAW,EACX,YAAY,EACZ,kBAAkB,GACrB,MAAM,mBAAmB,CAAC;AAE3B,OAAO,EAAE,kBAAkB,EAAE,MAAM,4BAA4B,CAAC;AAIhE,eAAO,MAAM,OAAO,UAAc,CAAC;AACnC,eAAO,MAAM,WAAW,wFAAwF,CAAC"}
|
package/dist/index.js
CHANGED
|
@@ -10,6 +10,11 @@ export { Mamba3Block } from './model/mamba3_block.js';
|
|
|
10
10
|
export { AttentionBlock } from './model/attention_block.js';
|
|
11
11
|
// Deprecated alias — kept until 3.0.0
|
|
12
12
|
export { MambaBlock } from './model/mamba1_block.js';
|
|
13
|
+
// ── Mixture-of-Experts (shared-expert hybrid sparsity) ─────────────────────────
|
|
14
|
+
export { SharedExpertMoE, LoadBalanceAccumulator, DEFAULT_MOE_CONFIG, DEFAULT_MOE_SEED, MoETrainer, EvermindModelPackage, } from './moe/index.js';
|
|
15
|
+
// ── EvermindLM (the generative model) + AdamW ──────────────────────────────────
|
|
16
|
+
export { EvermindLM, EvermindLMTrainer, DEFAULT_LM_CONFIG, DEFAULT_LM_SEED } from './lm/index.js';
|
|
17
|
+
export { AdamW } from './optim/adamw.js';
|
|
13
18
|
// ── Training ──────────────────────────────────────────────────────────────────
|
|
14
19
|
export { MambaTrainer } from './training/trainer.js';
|
|
15
20
|
export { Tensor, backward, enableGrad, noGrad, clearTape, recordOperation, crossEntropyLoss, crossEntropyGrad, } from './training/autograd.js';
|
package/dist/index.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,iFAAiF;AAEjF,OAAO,EAAE,gBAAgB,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAEtE,oBAAoB;AACpB,OAAO,EAAE,WAAW,EAAE,MAAQ,yBAAyB,CAAC;AACxD,OAAO,EAAE,WAAW,EAAE,MAAQ,yBAAyB,CAAC;AACxD,OAAO,EAAE,WAAW,EAAE,MAAQ,yBAAyB,CAAC;AACxD,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAE5D,sCAAsC;AACtC,OAAO,EAAE,UAAU,EAAE,MAAM,yBAAyB,CAAC;AAErD,iFAAiF;AAEjF,OAAO,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAC;AACrD,OAAO,EACH,MAAM,EACN,QAAQ,EACR,UAAU,EACV,MAAM,EACN,SAAS,EACT,eAAe,EACf,gBAAgB,EAChB,gBAAgB,GACnB,MAAM,wBAAwB,CAAC;AAEhC,iFAAiF;AAEjF,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAElD,iFAAiF;AAEjF,OAAO,EAAE,SAAS,EAAE,WAAW,EAAE,KAAK,EAAE,aAAa,EAAE,MAAM,gBAAgB,CAAC;AAkB9E,iFAAiF;AAEjF,OAAO,EACH,UAAU,EACV,mBAAmB,EACnB,wBAAwB,EACxB,mBAAmB,EACnB,qBAAqB,EACrB,eAAe,EACf,cAAc,EACd,UAAU,EACV,YAAY,EACZ,IAAI,GACP,MAAM,sBAAsB,CAAC;AAE9B,iFAAiF;AAEjF,OAAO,EACH,YAAY,EACZ,cAAc,EACd,WAAW,EACX,WAAW,EACX,YAAY,EACZ,cAAc,EACd,sBAAsB,EACtB,wBAAwB,EACxB,cAAc,GACjB,MAAM,yBAAyB,CAAC;AAEjC,iFAAiF;AAEjF,8BAA8B;AAC9B,OAAO,EAAE,2BAA2B,EAAE,4BAA4B,EAAE,MAC3D,6BAA6B,CAAC;AACvC,OAAO,EAAE,mBAAmB,EAAE,oBAAoB,EAAE,MAC3C,qBAAqB,CAAC;AAC/B,OAAO,EAAE,mBAAmB,EAAE,oBAAoB,EAAE,MAC3C,gCAAgC,CAAC;AAC1C,OAAO,EAAE,kBAAkB,EAAE,cAAc,EAAE,MACpC,4BAA4B,CAAC;AACtC,OAAO,EAAE,gBAAgB,EAAE,yBAAyB,EAAE,oBAAoB,EAAE,qBAAqB,EAAE,MAC1F,0BAA0B,CAAC;AAEpC,sBAAsB;AACtB,OAAO,EAAE,gBAAgB,EAAE,iBAAiB,EAAE,MACrC,kBAAkB,CAAC;AAE5B,8BAA8B;AAC9B,OAAO,EAAE,wBAAwB,EAAE,yBAAyB,EAAE,MACrD,0BAA0B,CAAC;AAEpC,oBAAoB;AACpB,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,YAAY,EAAE,MAC/D,wBAAwB,CAAC;AAElC,iFAAiF;AAEjF,OAAO,EACH,MAAM,EACN,UAAU,EACV,gBAAgB,EAChB,gBAAgB,EAChB,aAAa,EACb,aAAa,EACb,QAAQ,EACR,UAAU,EACV,YAAY,EACZ,aAAa,EACb,aAAa,EACb,WAAW,EACX,qBAAqB,EACrB,mBAAmB,EACnB,aAAa,GAChB,MAAM,mBAAmB,CAAC;AAW3B,OAAO,EAAE,kBAAkB,EAAE,MAAM,4BAA4B,CAAC;AAEhE,iFAAiF;AAEjF,MAAM,CAAC,MAAM,OAAO,GAAO,OAAO,CAAC;AACnC,MAAM,CAAC,MAAM,WAAW,GAAG,qFAAqF,CAAC"}
|
|
1
|
+
{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,iFAAiF;AAEjF,OAAO,EAAE,gBAAgB,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAEtE,oBAAoB;AACpB,OAAO,EAAE,WAAW,EAAE,MAAQ,yBAAyB,CAAC;AACxD,OAAO,EAAE,WAAW,EAAE,MAAQ,yBAAyB,CAAC;AACxD,OAAO,EAAE,WAAW,EAAE,MAAQ,yBAAyB,CAAC;AACxD,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAE5D,sCAAsC;AACtC,OAAO,EAAE,UAAU,EAAE,MAAM,yBAAyB,CAAC;AAErD,kFAAkF;AAClF,OAAO,EACH,eAAe,EACf,sBAAsB,EACtB,kBAAkB,EAClB,gBAAgB,EAChB,UAAU,EACV,oBAAoB,GACvB,MAAM,gBAAgB,CAAC;AAcxB,kFAAkF;AAClF,OAAO,EAAE,UAAU,EAAE,iBAAiB,EAAE,iBAAiB,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAElG,OAAO,EAAE,KAAK,EAAE,MAAM,kBAAkB,CAAC;AAGzC,iFAAiF;AAEjF,OAAO,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAC;AACrD,OAAO,EACH,MAAM,EACN,QAAQ,EACR,UAAU,EACV,MAAM,EACN,SAAS,EACT,eAAe,EACf,gBAAgB,EAChB,gBAAgB,GACnB,MAAM,wBAAwB,CAAC;AAEhC,iFAAiF;AAEjF,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAElD,iFAAiF;AAEjF,OAAO,EAAE,SAAS,EAAE,WAAW,EAAE,KAAK,EAAE,aAAa,EAAE,MAAM,gBAAgB,CAAC;AAkB9E,iFAAiF;AAEjF,OAAO,EACH,UAAU,EACV,mBAAmB,EACnB,wBAAwB,EACxB,mBAAmB,EACnB,qBAAqB,EACrB,eAAe,EACf,cAAc,EACd,UAAU,EACV,YAAY,EACZ,IAAI,GACP,MAAM,sBAAsB,CAAC;AAE9B,iFAAiF;AAEjF,OAAO,EACH,YAAY,EACZ,cAAc,EACd,WAAW,EACX,WAAW,EACX,YAAY,EACZ,cAAc,EACd,sBAAsB,EACtB,wBAAwB,EACxB,cAAc,GACjB,MAAM,yBAAyB,CAAC;AAEjC,iFAAiF;AAEjF,8BAA8B;AAC9B,OAAO,EAAE,2BAA2B,EAAE,4BAA4B,EAAE,MAC3D,6BAA6B,CAAC;AACvC,OAAO,EAAE,mBAAmB,EAAE,oBAAoB,EAAE,MAC3C,qBAAqB,CAAC;AAC/B,OAAO,EAAE,mBAAmB,EAAE,oBAAoB,EAAE,MAC3C,gCAAgC,CAAC;AAC1C,OAAO,EAAE,kBAAkB,EAAE,cAAc,EAAE,MACpC,4BAA4B,CAAC;AACtC,OAAO,EAAE,gBAAgB,EAAE,yBAAyB,EAAE,oBAAoB,EAAE,qBAAqB,EAAE,MAC1F,0BAA0B,CAAC;AAEpC,sBAAsB;AACtB,OAAO,EAAE,gBAAgB,EAAE,iBAAiB,EAAE,MACrC,kBAAkB,CAAC;AAE5B,8BAA8B;AAC9B,OAAO,EAAE,wBAAwB,EAAE,yBAAyB,EAAE,MACrD,0BAA0B,CAAC;AAEpC,oBAAoB;AACpB,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,YAAY,EAAE,MAC/D,wBAAwB,CAAC;AAElC,iFAAiF;AAEjF,OAAO,EACH,MAAM,EACN,UAAU,EACV,gBAAgB,EAChB,gBAAgB,EAChB,aAAa,EACb,aAAa,EACb,QAAQ,EACR,UAAU,EACV,YAAY,EACZ,aAAa,EACb,aAAa,EACb,WAAW,EACX,qBAAqB,EACrB,mBAAmB,EACnB,aAAa,GAChB,MAAM,mBAAmB,CAAC;AAW3B,OAAO,EAAE,kBAAkB,EAAE,MAAM,4BAA4B,CAAC;AAEhE,iFAAiF;AAEjF,MAAM,CAAC,MAAM,OAAO,GAAO,OAAO,CAAC;AACnC,MAAM,CAAC,MAAM,WAAW,GAAG,qFAAqF,CAAC"}
|
|
@@ -0,0 +1,148 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* evermind_lm.ts — EvermindLM: a small but complete generative language model.
|
|
3
|
+
*
|
|
4
|
+
* This is what turns a trained checkpoint into an *AI that generates text* (the
|
|
5
|
+
* thing a marketplace buyer actually runs). Architecture (Mamba-flavoured, the
|
|
6
|
+
* minimal exact-gradient CPU reference):
|
|
7
|
+
*
|
|
8
|
+
* x_t = Embed[token_t]
|
|
9
|
+
* per layer:
|
|
10
|
+
* x_t += DepthwiseCausalConv(x)_t // temporal mixing (short conv)
|
|
11
|
+
* x_t += SharedExpertMoE(x_t) // per-position channel mixing (sparse)
|
|
12
|
+
* logits_t = x_t · Embedᵀ // tied output head
|
|
13
|
+
*
|
|
14
|
+
* The token mixer is a depthwise causal convolution (each channel sees a short
|
|
15
|
+
* window of its own past — Mamba's pre-conv) and the channel mixer is the
|
|
16
|
+
* shared-expert MoE, so the model is genuinely sparse. Embeddings are tied
|
|
17
|
+
* (input lookup == output head), which the gradient code accounts for.
|
|
18
|
+
*
|
|
19
|
+
* Pure CPU, exact forward + backward (finite-difference checked), reusing the
|
|
20
|
+
* engine's MoE, cross-entropy, and AdamW. The WGSL/WebGPU path is a future
|
|
21
|
+
* acceleration with the same shapes.
|
|
22
|
+
*/
|
|
23
|
+
import { type AdamWOptions } from "../optim/adamw.js";
|
|
24
|
+
export interface EvermindLMConfig {
|
|
25
|
+
/** Vocabulary size. */
|
|
26
|
+
vocabSize: number;
|
|
27
|
+
/** Model (channel) dimension. Default 64. */
|
|
28
|
+
dModel: number;
|
|
29
|
+
/** Number of (conv + MoE) blocks. Default 2. */
|
|
30
|
+
numLayers: number;
|
|
31
|
+
/** Causal conv kernel width. Default 3. */
|
|
32
|
+
convKernel: number;
|
|
33
|
+
/** Hidden width of each MoE expert FFN. Default 2·dModel. */
|
|
34
|
+
hiddenDim: number;
|
|
35
|
+
/** Routed experts per MoE layer. Default 4. */
|
|
36
|
+
numExperts: number;
|
|
37
|
+
/** Experts activated per token. Default 2. */
|
|
38
|
+
topK: number;
|
|
39
|
+
/** Deterministic init seed. */
|
|
40
|
+
seed?: number;
|
|
41
|
+
}
|
|
42
|
+
export declare const DEFAULT_LM_CONFIG: Required<Omit<EvermindLMConfig, "seed" | "vocabSize">>;
|
|
43
|
+
export declare const DEFAULT_LM_SEED = 1163283533;
|
|
44
|
+
interface MoECacheLike {
|
|
45
|
+
x: Float32Array;
|
|
46
|
+
route: {
|
|
47
|
+
experts: number[];
|
|
48
|
+
gates: number[];
|
|
49
|
+
probs: Float32Array;
|
|
50
|
+
};
|
|
51
|
+
sharedPre: Float32Array;
|
|
52
|
+
sharedH: Float32Array;
|
|
53
|
+
expertOut: Float32Array[];
|
|
54
|
+
expertPre: Float32Array[];
|
|
55
|
+
expertH: Float32Array[];
|
|
56
|
+
}
|
|
57
|
+
interface LayerCache {
|
|
58
|
+
layerIn: Float32Array[];
|
|
59
|
+
normedConv: Float32Array[];
|
|
60
|
+
rmsConv: number[];
|
|
61
|
+
afterConv: Float32Array[];
|
|
62
|
+
rmsMoe: number[];
|
|
63
|
+
moeCache: MoECacheLike[];
|
|
64
|
+
}
|
|
65
|
+
interface ForwardCache {
|
|
66
|
+
tokens: number[];
|
|
67
|
+
layers: LayerCache[];
|
|
68
|
+
finalX: Float32Array[];
|
|
69
|
+
}
|
|
70
|
+
/** A tokenizer the LM can read/write text through (the engine's `BPETokenizer` fits). */
|
|
71
|
+
export interface TextCodec {
|
|
72
|
+
encode(text: string): number[];
|
|
73
|
+
decode(ids: number[]): string;
|
|
74
|
+
}
|
|
75
|
+
export interface LMGenerateOptions {
|
|
76
|
+
maxNewTokens: number;
|
|
77
|
+
/** Sampling temperature; ≤0 ⇒ greedy argmax. Default 0 (greedy). */
|
|
78
|
+
temperature?: number;
|
|
79
|
+
/** Deterministic sampler seed (only used when temperature > 0). */
|
|
80
|
+
seed?: number;
|
|
81
|
+
/** Stop generating when this token id is produced. */
|
|
82
|
+
stopToken?: number;
|
|
83
|
+
}
|
|
84
|
+
export declare class EvermindLM {
|
|
85
|
+
readonly config: Required<Omit<EvermindLMConfig, "seed">>;
|
|
86
|
+
/** Tied token embedding / output head: vocabSize × dModel (row-major). */
|
|
87
|
+
emb: Float32Array;
|
|
88
|
+
private gEmb;
|
|
89
|
+
/** Per-layer depthwise causal conv kernels: dModel × convKernel. */
|
|
90
|
+
private readonly conv;
|
|
91
|
+
private readonly gConv;
|
|
92
|
+
/** Per-layer pre-conv / pre-MoE RMSNorm gains (dModel each). */
|
|
93
|
+
private readonly nConv;
|
|
94
|
+
private readonly gNConv;
|
|
95
|
+
private readonly nMoe;
|
|
96
|
+
private readonly gNMoe;
|
|
97
|
+
/** Per-layer channel mixer. */
|
|
98
|
+
private readonly moe;
|
|
99
|
+
constructor(config: EvermindLMConfig);
|
|
100
|
+
/** Run the model over a token sequence; returns per-position logits + a cache. */
|
|
101
|
+
forward(tokens: number[]): {
|
|
102
|
+
logits: Float32Array[];
|
|
103
|
+
cache: ForwardCache;
|
|
104
|
+
};
|
|
105
|
+
/**
|
|
106
|
+
* Next-token cross-entropy over the sequence (predict tokens[t+1] from
|
|
107
|
+
* position t), accumulating exact gradients. Returns the mean loss. Call
|
|
108
|
+
* {@link zeroGrad} before and an optimiser step after.
|
|
109
|
+
*/
|
|
110
|
+
lossAndBackward(tokens: number[]): number;
|
|
111
|
+
/**
|
|
112
|
+
* Text-level generation: encode the prompt, generate, decode. `codec` is any
|
|
113
|
+
* tokenizer exposing encode/decode (the engine's `BPETokenizer` satisfies it),
|
|
114
|
+
* so the LM consumes and emits real text rather than raw token ids. The model's
|
|
115
|
+
* `vocabSize` must match the codec's vocabulary.
|
|
116
|
+
*/
|
|
117
|
+
generateText(prompt: string, codec: TextCodec, opts: LMGenerateOptions): string;
|
|
118
|
+
/** Greedy / temperature-sampled autoregressive generation. Returns NEW token ids. */
|
|
119
|
+
generate(prompt: number[], opts: LMGenerateOptions): number[];
|
|
120
|
+
/** All trainable parameters as {data} (AdamW-compatible), canonical order. */
|
|
121
|
+
parameters(): {
|
|
122
|
+
data: Float32Array;
|
|
123
|
+
}[];
|
|
124
|
+
/** Gradient buffers, index-aligned with {@link parameters}. */
|
|
125
|
+
gradients(): {
|
|
126
|
+
data: Float32Array;
|
|
127
|
+
}[];
|
|
128
|
+
zeroGrad(): void;
|
|
129
|
+
/** Serialise to an "EVL0" binary (fp16 or f32), params in {@link parameters} order. */
|
|
130
|
+
exportWeights(opts?: {
|
|
131
|
+
fp16?: boolean;
|
|
132
|
+
}): ArrayBuffer;
|
|
133
|
+
/** Load weights from an "EVL0" binary. Validates magic + dims. */
|
|
134
|
+
loadWeights(buffer: ArrayBuffer): void;
|
|
135
|
+
}
|
|
136
|
+
/** Minimal sequence trainer: AdamW over next-token cross-entropy. */
|
|
137
|
+
export declare class EvermindLMTrainer {
|
|
138
|
+
private readonly model;
|
|
139
|
+
private readonly opts;
|
|
140
|
+
private readonly adam;
|
|
141
|
+
constructor(model: EvermindLM, opts?: AdamWOptions & {
|
|
142
|
+
epochs?: number;
|
|
143
|
+
});
|
|
144
|
+
/** Train on a set of token sequences; returns per-epoch mean loss. */
|
|
145
|
+
fit(sequences: number[][]): number[];
|
|
146
|
+
}
|
|
147
|
+
export {};
|
|
148
|
+
//# sourceMappingURL=evermind_lm.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evermind_lm.d.ts","sourceRoot":"","sources":["../../src/lm/evermind_lm.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;GAqBG;AAIH,OAAO,EAAS,KAAK,YAAY,EAAE,MAAM,mBAAmB,CAAC;AAI7D,MAAM,WAAW,gBAAgB;IAC/B,uBAAuB;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,6CAA6C;IAC7C,MAAM,EAAE,MAAM,CAAC;IACf,gDAAgD;IAChD,SAAS,EAAE,MAAM,CAAC;IAClB,2CAA2C;IAC3C,UAAU,EAAE,MAAM,CAAC;IACnB,6DAA6D;IAC7D,SAAS,EAAE,MAAM,CAAC;IAClB,+CAA+C;IAC/C,UAAU,EAAE,MAAM,CAAC;IACnB,8CAA8C;IAC9C,IAAI,EAAE,MAAM,CAAC;IACb,+BAA+B;IAC/B,IAAI,CAAC,EAAE,MAAM,CAAC;CACf;AAED,eAAO,MAAM,iBAAiB,EAAE,QAAQ,CAAC,IAAI,CAAC,gBAAgB,EAAE,MAAM,GAAG,WAAW,CAAC,CAOpF,CAAC;AAEF,eAAO,MAAM,eAAe,aAAa,CAAC;AAG1C,UAAU,YAAY;IACpB,CAAC,EAAE,YAAY,CAAC;IAChB,KAAK,EAAE;QAAE,OAAO,EAAE,MAAM,EAAE,CAAC;QAAC,KAAK,EAAE,MAAM,EAAE,CAAC;QAAC,KAAK,EAAE,YAAY,CAAA;KAAE,CAAC;IACnE,SAAS,EAAE,YAAY,CAAC;IACxB,OAAO,EAAE,YAAY,CAAC;IACtB,SAAS,EAAE,YAAY,EAAE,CAAC;IAC1B,SAAS,EAAE,YAAY,EAAE,CAAC;IAC1B,OAAO,EAAE,YAAY,EAAE,CAAC;CACzB;AAED,UAAU,UAAU;IAClB,OAAO,EAAE,YAAY,EAAE,CAAC;IACxB,UAAU,EAAE,YAAY,EAAE,CAAC;IAC3B,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,SAAS,EAAE,YAAY,EAAE,CAAC;IAC1B,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,QAAQ,EAAE,YAAY,EAAE,CAAC;CAC1B;AAED,UAAU,YAAY;IACpB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,MAAM,EAAE,UAAU,EAAE,CAAC;IACrB,MAAM,EAAE,YAAY,EAAE,CAAC;CACxB;AAED,yFAAyF;AACzF,MAAM,WAAW,SAAS;IACxB,MAAM,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,CAAC;IAC/B,MAAM,CAAC,GAAG,EAAE,MAAM,EAAE,GAAG,MAAM,CAAC;CAC/B;AAED,MAAM,WAAW,iBAAiB;IAChC,YAAY,EAAE,MAAM,CAAC;IACrB,oEAAoE;IACpE,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,mEAAmE;IACnE,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,sDAAsD;IACtD,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,qBAAa,UAAU;IACrB,QAAQ,CAAC,MAAM,EAAE,QAAQ,CAAC,IAAI,CAAC,gBAAgB,EAAE,MAAM,CAAC,CAAC,CAAC;IAE1D,0EAA0E;IAC1E,GAAG,EAAE,YAAY,CAAC;IAClB,OAAO,CAAC,IAAI,CAAe;IAC3B,oEAAoE;IACpE,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAiB;IACtC,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAiB;IACvC,gEAAgE;IAChE,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAiB;IACvC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAiB;IACxC,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAiB;IACtC,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAiB;IACvC,+BAA+B;IAC/B,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAoB;gBAE5B,MAAM,EAAE,gBAAgB;IA+DpC,kFAAkF;IAClF,OAAO,CAAC,MAAM,EAAE,MAAM,EAAE,GAAG;QAAE,MAAM,EAAE,YAAY,EAAE,CAAC;QAAC,KAAK,EAAE,YAAY,CAAA;KAAE;IA4E1E;;;;OAIG;IACH,eAAe,CAAC,MAAM,EAAE,MAAM,EAAE,GAAG,MAAM;IAqFzC;;;;;OAKG;IACH,YAAY,CAAC,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,EAAE,iBAAiB,GAAG,MAAM;IAI/E,qFAAqF;IACrF,QAAQ,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,IAAI,EAAE,iBAAiB,GAAG,MAAM,EAAE;IAkB7D,8EAA8E;IAC9E,UAAU,IAAI;QAAE,IAAI,EAAE,YAAY,CAAA;KAAE,EAAE;IAStC,+DAA+D;IAC/D,SAAS,IAAI;QAAE,IAAI,EAAE,YAAY,CAAA;KAAE,EAAE;IASrC,QAAQ,IAAI,IAAI;IAUhB,uFAAuF;IACvF,aAAa,CAAC,IAAI,GAAE;QAAE,IAAI,CAAC,EAAE,OAAO,CAAA;KAAO,GAAG,WAAW;IA+BzD,kEAAkE;IAClE,WAAW,CAAC,MAAM,EAAE,WAAW,GAAG,IAAI;CA4BvC;AAED,qEAAqE;AACrE,qBAAa,iBAAiB;IAG1B,OAAO,CAAC,QAAQ,CAAC,KAAK;IACtB,OAAO,CAAC,QAAQ,CAAC,IAAI;IAHvB,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAQ;gBAEV,KAAK,EAAE,UAAU,EACjB,IAAI,GAAE,YAAY,GAAG;QAAE,MAAM,CAAC,EAAE,MAAM,CAAA;KAAO;IAIhE,sEAAsE;IACtE,GAAG,CAAC,SAAS,EAAE,MAAM,EAAE,EAAE,GAAG,MAAM,EAAE;CAiBrC"}
|