Repository Issue Activity (beta)

#148 Request: Bonsai conversion of ThinkingCap-Qwen3.6-27B (halves thinking tokens, composes with ternary)

#147 [27B multimodal] Repeated 14K prompts get no prefix-cache reuse

#137 Please consider making a mainline compatible dspark draft model or publish some advice on that

#146 [27B/OpenAI API] Clarify reasoning defaults and per-request opt-out for agentic tool calls

#145 [CUDA/mainline] Q4 KV causes a 3x-67x prompt-processing slowdown on Ternary-Bonsai-27B

#81 GLM 5.2 version ?

#142 Just comment: bonsai 1.7B on haswell i5 iMac

#92 Ternary-Bonsai-27B on ROCm / RDNA3: works unpatched (RX 7800 XT, gfx1101) + benchmarks

#102 Together serverless Ternary Bonsai 27B endpoint metadata points at an AWQ-4bit artifact, catalog says ternary g128

#139 Combination of BONSAI_KV4 and BONSAI_SPECULATIVE on 27B Ternary reports issue with KV cache tpye

#114 One-click Windows launcher for Bonsai-demo (community tool, not a PR)

#131 It is stuck in a loop when used with openclaw model Ternary-Bonsai-27B-Q2_0

#130 Bonsai Alternatives

#101 Any plans to release the ternary vLLM kernels used in the whitepaper evals?

#125 --no-mmproj-offload should be specified for vision

#105 DSpark drafter: measured 32k enablement ceiling + net-negative speedup at depth (mechanism ≠ VRAM)

#121 Chance of Qwen 3.8 and/or KIMI K3?

#124 Context should not be specified with `-c 0` otherwise fitting fails.

#86 Ornith-1.0 9B version?

#115 bug(deps): open-webui 0.10.2 routers/audio requires Python *strictly* <3.13 (or traceback)

#93 Pre-built llama.cpp binaries: Metal fails to compile on macOS 26 Tahoe (M5)

#112 Speculative decoding (draft-dspark) silently falls back to CPU on DGX Spark / unified-memory CUDA devices — 2.35x slower than fixed

#85 is iphone's NPU utilized in Bonsai inference?

#113 Reverts to GPU inference when running it on AMD CPU with integrated graphics (only affects llama server)

#95 Question: Exact PrismML-MLX commit used for Bonsai-27B-mlx-1bit release