@tryhamster/gerbil 1.1.0 → 1.1.1

package/dist/{gpu-CUOj5CaA.mjs → gpu-yxEq7qT1.mjs}

@@ -1,6 +1,6 @@
 import { n as resolveDefaultRepo } from "./defaults-9komdrbY.mjs";
 import { C as parseKaniConfig, E as GEMMA4_VIS_KEYS, T as DTYPE_BYTES, _ as generateNanoCodecDecoderGraph, b as kaniLayerAlpha, d as KANI_START_OF_HUMAN, f as audioTokensToCodes, g as generateKaniTtsGraph, m as computeKaniPositions, p as buildKaniLayerCosSin, u as KANI_END_OF_HUMAN, v as kaniAttentionLayerIndices, w as CANONICAL_KEYS, x as kaniSinTensor, y as kaniCosTensor } from "./architectures-C1I5V3Dt.mjs";
-import { _ as verifyGPU, c as KERNEL_REGISTRY, d as createBindGroup, f as createStorageBuffer, g as initGPU, h as getOrCreatePipeline, i as loadModel, l as MATMUL_BIAS_F16C_SPEC, m as destroyBuffers, o as remapPrunedToken, p as createUniformBuffer, r as loadKaniTTS, s as Executor, u as clearPipelineCache } from "./moonshine-stt-aSSceh2Y.mjs";
+import { _ as verifyGPU, c as KERNEL_REGISTRY, d as createBindGroup, f as createStorageBuffer, g as initGPU, h as getOrCreatePipeline, i as loadModel, l as MATMUL_BIAS_F16C_SPEC, m as destroyBuffers, o as remapPrunedToken, p as createUniformBuffer, r as loadKaniTTS, s as Executor, u as clearPipelineCache } from "./moonshine-stt-Cqdf1myz.mjs";
 
 //#region src/gpu/architectures/gemma4_vision.ts
 /**
@@ -2845,8 +2845,8 @@ var WebGPUEngine = class WebGPUEngine {
 		let mmLogicalPos = inputIds.length;
 		const keepPlan = this.config.vocabKeepPlan;
 		const remapTok = keepPlan ? (i) => remapPrunedToken(i, keepPlan) : (i) => i;
-		if (isGreedy && !this.executor.needsMultiEncoder && !mmDecode && !keepPlan) {
-			const firstToken = sampleToken(logits, sampling, [...inputIds, ...generatedIds]);
+		if (isGreedy && !this.executor.needsMultiEncoder && !mmDecode) {
+			const firstToken = remapTok(sampleToken(logits, sampling, [...inputIds, ...generatedIds]));
 			if (!consumeToken(firstToken)) {
 				const depth = Executor.PIPELINE_DEPTH;
 				const stepsNeeded = Math.min(maxTokens - 1, this.executor.decodeCapacityRemaining());
@@ -2867,7 +2867,7 @@ var WebGPUEngine = class WebGPUEngine {
 			if (step === 0 || !isGreedy) nextToken = remapTok(sampleToken(logits, sampling, [...inputIds, ...generatedIds]));
 			else {
 				if (mmDecode) this.writeMRoPEDecodeStep(this.executor.currentSeqPos, mmLogicalPos);
-				nextToken = remapTok(await this.executor.forwardArgmax(new Uint32Array([generatedIds[generatedIds.length - 1]])));
+				nextToken = await this.executor.forwardArgmax(new Uint32Array([generatedIds[generatedIds.length - 1]]));
 				mmLogicalPos++;
 			}
 			if (consumeToken(nextToken)) break;
@@ -3799,4 +3799,4 @@ var WebGPUEngine = class WebGPUEngine {
 
 //#endregion
 export { generateGemma4VisionGraph as C, dequantizeMLXProjection as S, resolveGemma4VisionInfo as T, smartResize as _, buildGemma4PosEmbeds as a, generateQwen3_5VisionGraph as b, buildMRoPECosSin as c, buildPositionIds as d, buildRotaryCosSin as f, preprocessImageGemma4 as g, preprocessImage as h, buildGemma4PoolMatrix as i, buildMRoPEPositionIds as l, mropeFreqDims as m, GEMMA4_IMAGE_PROCESSOR as n, buildGemma4RotaryCosSin as o, buildVisionPositionTensors as p, QWEN3_5_IMAGE_PROCESSOR as r, buildGemma4VisionPositionTensors as s, WebGPUEngine as t, buildPosEmbeds as u, VisionExecutor as v, patchGemma4VisionClips as w, dequantizeGemma4VisionProjection as x, KaniTTS as y };
-//# sourceMappingURL=gpu-CUOj5CaA.mjs.map
+//# sourceMappingURL=gpu-yxEq7qT1.mjs.map