Repository navigation
Restore upstream runtime with Eliza model and voice compatibility - #72
Merged
Merged
Conversation
github-actions
Bot
added
documentation
lalalune
marked this pull request as ready for review
October 14, 2026 04:55
Member
Author
|
FYI from the pin-branch work: #77 merges 🤖 Generated with Claude Code |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Preserve the upstream runtime update alongside the Eliza voice and custom-quantization integration from #65 and #66. The merged main retained the older runtime; this follow-up restores upstream model/backend changes and explicitly ports the Eliza functionality onto them.
Upstream Q2_0 keeps type ID 42. Eliza types move to 256–264, with native/Python legacy readers and an explicit migration tool for ambiguous old files. Restore custom backend registrations, adapt Metal/Vulkan/CUDA integration, preserve dynamic/shared KV caches, and update multimodal FFI calls for the upstream bitmap and prompt-length APIs. Include the speech type-check and Metal workflow fixes subsequently merged into main.
Validation completed: Python GGUF tests (8); native GGUF compatibility, KV growth, WordPiece normalization, fused ABI and Paradee ASR contract tests; Kokoro portable layers and Paradee phase lock; CPU quantization; Metal Q2_0 parity (124), custom types (21), custom attention/ISTFT (16), capability and guarded shared/private readback; Vulkan parity (111 supported cases); RPC and threaded WASM/WebGPU builds. A real Qwen 2B model generated 32 tokens through Metal. Eliza's Bun FFI loader opened the actual ABI 18 library successfully.
Vulkan general custom-type copy/matmul and device-gated Metal TBQ flash attention report unsupported and are not counted as passes. CUDA hardware and browser WebGPU execution are unverified on this host. Hosted Actions are waived as a merge gate by the maintainer.
Remove the stale tracked web UI node_modules tree inherited from the re-root. Preserve the newly merged voice fixes and add regression coverage for independent WordPiece lowercase/accent settings.