Block a user
Add a performance-parity Qwen3.8-27B Q4 and Q8 text backend on Metal
Based on findings in #84, all work on MOJO migration is stopped. Licene-hell.
Implement the complete DeepSeek Mojo-Metal backend with performance parity
Implement the complete DeepSeek Mojo-Metal backend with performance parity
Based on findings in #84, all work on MOJO migration is stopped. Licene-hell.
Implement the shared Mojo foundation and a quantized Metal vertical slice
Implement the shared Mojo foundation and a quantized Metal vertical slice
Based on findings in #84, all work on MOJO migration is stopped. Licene-hell.
Build the native-Metal-versus-Mojo parity harness and operation inventory
Build the native-Metal-versus-Mojo parity harness and operation inventory
Based on findings in #84, all work on MOJO migration is stopped. Licene-hell.
Spike: prove a license-safe, performance-parity Mojo-Metal integration
Spike: prove a license-safe, performance-parity Mojo-Metal integration
Spike: prove a license-safe, performance-parity Mojo-Metal integration
based on this findings - especially the license topics - all work on MOJO is cancelled.
Spike: prove a license-safe, performance-parity Mojo-Metal integration
Spike: prove a license-safe, performance-parity Mojo-Metal integration
Implemented and pushed in 0dcf181.
Decision: STOP at the mandatory license/runtime gate. Mojo is pinned to 1.0.0 (Mojo 1.0.0 (ed45d567)) through uv.lock, with a one-command proof at…
Optional follow-up: add CUDA to the completed Mojo backend
Complete the Mojo release matrix, packaging, cutover, and native-backend retirement
Implement the GLM 5.2 Mojo backend with isolated sparse-path parity gates
Add Qwen3.8 MTP and multimodal inference without regressing the text path
Add a performance-parity Qwen3.8-27B Q4 and Q8 text backend in Mojo