Issues

• InfiniteTalk

  apple/coreai-models#343 · Filter by author stikves opened on Oct 11, 2026

• Streaming detokenization is O(N²): it re-decodes the whole token sequence every step

  apple/coreai-models#341 · Filter by author stikves opened on Oct 11, 2026

• Aleph-Alpha/Kolibri-1

  apple/coreai-models#330 · Filter by author stikves opened on Oct 6, 2026

• Model request: Speech enhancement — RNNoise, GTCRN and UL-UNAS

  apple/coreai-models#325 · Filter by author maxffarrell opened on Oct 5, 2026

• Model request: Julia-1 decision model for macOS

  apple/coreai-models#324 · Filter by author maxffarrell opened on Oct 5, 2026

• [Swift runtime] Pipelined constrained generation binds only the K/V pair, so bundles with more states fail with "Missing state view"

  apple/coreai-models#316 · Filter by author iannellomarco opened on Oct 1, 2026

• [Swift runtime] Streamed text re-emits text when a token extends the previous grapheme (ZWJ emoji, combining marks, Indic signs)

  apple/coreai-models#315 · Filter by author iannellomarco opened on Oct 1, 2026

• [macOS export] RingKVCache with capacity == sliding_window is wrong for chunked prefill once the ring is full (Muse Glimmer default chunking)

  apple/coreai-models#314 · Filter by author iannellomarco opened on Oct 1, 2026

• Core AI GPU gather from a constant with 2^31 or more elements silently returns wrong rows; exporter should warn or split (affects gemma-3n-E4B per-layer embeddings)

  apple/coreai-models#313 · Filter by author iannellomarco opened on Oct 1, 2026

• [Swift runtime] A failed GPU command buffer is swallowed: the call returns normally and the pipelined engine streams <pad> tokens

  apple/coreai-models#312 · Filter by author iannellomarco opened on Oct 1, 2026

• Sliding-window SDPA (is_causal + window_size) on strided K/V views returns NaN for a short chunk over more than 3,584 keys in deep dynamic graphs

  apple/coreai-models#311 · Filter by author iannellomarco opened on Oct 1, 2026

• [Python runtime] The 7th distinct input shape of a dynamic function fails (MTL4CommandQueueErrorDomain error 1), returns zeros, and the function stays broken

  apple/coreai-models#310 · Filter by author iannellomarco opened on Oct 1, 2026

• [Swift runtime] Guided generation with Qwen3-4B repeats until maximumResponseTokens; timeout cancellation not honoured (experimental export)

  apple/coreai-models#300 · Filter by author emusickid opened on Sep 29, 2026

• Tag a release that builds with Xcode 27

  bug
  apple/coreai-models#293 · Filter by author willswire opened on Sep 26, 2026

• Transformers version

  apple/coreai-models#292 · Filter by author kevchengcodes opened on Sep 26, 2026

• public enum StateKind has incorrect documentation

  code health
  apple/coreai-models#288 · Filter by author stikves opened on Sep 24, 2026

• Explore Video Diffusion VAE tile size

  bug
  apple/coreai-models#285 · Filter by author kevchengcodes opened on Sep 22, 2026

• The RoPE cos/sin gather composite allows only one lookup per graph, fed straight from the position input

  apple/coreai-models#280 · Filter by author baozzz1 opened on Sep 22, 2026

• Per-block int weight quantization silently produces wrong output on the Neural Engine

  apple/coreai-models#279 · Filter by author baozzz1 opened on Sep 22, 2026
  1

• [Swift runtime] Sequential VLM engine rejects any decoder with more than two states

  apple/coreai-models#276 · Filter by author baozzz1 opened on Sep 22, 2026
  1

• Consolidate Random Number Generators

  apple/coreai-models#273 · Filter by author stikves opened on Sep 21, 2026

• Image segmentation flag and documentation

  apple/coreai-models#260 · Filter by author kevchengcodes opened on Sep 18, 2026
  1

  

• Static-shape (iOS) export produces corrupted output for custom hunyuan_v1_dense port — eager model is bit-exact.

  bug
  apple/coreai-models#257 · Filter by author ivanfioravanti opened on Sep 17, 2026

• ANECCompileOffline() failed: OSStatus=0, aneCompileStatus=1 with an empty ErrorList — coreai-build exits 0 and writes a GPU-only .aimodelc for a linear-INT4 static LLM (follow-up to #55)

  apple/coreai-models#205 · Filter by author john-rocky opened on Aug 27, 2026

• Failed to acquire the source buffer for the ViewOp — MPSGraph scratch heap overflows on any S>1 prefill of Gemma 4 E2B on iOS 27 (iOS face of #27, still on 24A5418b)

  apple/coreai-models#201 · Filter by author john-rocky opened on Aug 27, 2026
  1