Issues

When is the server version coming?
#206 · ghost opened on Apr 21, 2025
21

bitnet-b1.58-2B-4T: weight_scale precision (bf16 vs I2_S f32) and packed trits vs the bf16 master weights
microsoft/BitNet#632 · opened on Sep 23, 2026
1
build/bin/llama-cli throws SIGSEGV, Segmentation fault when using a TL2 GGUF model
microsoft/BitNet#631 · opened on Sep 16, 2026
2
Verify evals on Papers with Code
microsoft/BitNet#630 · opened on Sep 14, 2026
Latent int16 overflow in ggml_vec_dot_i2_i8_s_1x1: 128 maddubs results per lane against a 32767 ceiling
microsoft/BitNet#629 · opened on Sep 13, 2026
5
i2_s AVX2 GEMM folds the int16 accumulator 10x more often than its own bound requires (7.5% measured)
microsoft/BitNet#628 · opened on Sep 13, 2026
5
Regression: Falcon-E models cannot be loaded — "unknown pre-tokenizer type: 'falcon_e'" (support added in #268 was lost in the llama.cpp submodule update)
microsoft/BitNet#622 · opened on Sep 6, 2026
convert-hf-to-gguf-bitnet.py --outtype i2_s silently writes F16 (not I2_S) for LlamaForCausalLM BitNet checkpoints (Falcon3 / Falcon-E 1.58bit)
microsoft/BitNet#621 · opened on Sep 6, 2026
README "Convert from .safetensors" flow is broken: convert-ms-to-gguf-bitnet.py KeyError MODEL_ARCH.BITNET_25, and llama-quantize has no I2_S ftype (setup_env.py i2_s conversion path also affected)
microsoft/BitNet#619 · opened on Sep 5, 2026
ggml-cpu.c: src1_cont undeclared → build fails on ARM CPUs with i8mm (Apple M2/M3/M4, Graviton3, ...)
microsoft/BitNet#618 · opened on Sep 5, 2026
1
[Bug]: I2_S GEMM fast path produces garbage for multi-token prompts on AVX-only CPUs
microsoft/BitNet#617 · opened on Aug 26, 2026
1
setup_env.py on MacBook M2 failed
microsoft/BitNet#611 · opened on Aug 14, 2026
1
ARM i2_s (1.58-bit) CPU path produces constant output; fix in pinned llama.cpp submodule
microsoft/BitNet#610 · opened on Aug 11, 2026
2
Model Hallucination Occurs again and again.
microsoft/BitNet#609 · opened on Aug 11, 2026
Current bitnet-b1.58-2B-4T checkpoints contain all-zero MLP weight tensors (model generates garbage)
microsoft/BitNet#608 · opened on Aug 11, 2026
3
bitnet-b1.58-2B-4T runs FFN with SiLU instead of relu2 - wrong logits on every backend
microsoft/BitNet#602 · opened on Aug 8, 2026
2
i2_s SIGSEGVs at n_ubatch >= 32: BLAS backend dequantises by a row stride 4x the real packed row
microsoft/BitNet#601 · opened on Aug 6, 2026
2
BitNet-b1.58-2B-4T produces garbage output on ARM64/NEON (no AVX2) — scalar fallback uses wrong I2_S unpacking scheme
microsoft/BitNet#600 · opened on Aug 5, 2026
5
I2_S produces incorrect output on aarch64
microsoft/BitNet#598 · opened on Aug 3, 2026
[Windows] Linker error "undefined symbol: quantize_i2_s" when building with -q i2_s or -q tl2 on Windows 11
microsoft/BitNet#595 · opened on Jul 26, 2026
[Bug]: BitNet FFN uses SILU instead of ReLU² — perplexity 99.8 vs 17.1 on every CPU backend
microsoft/BitNet#588 · opened on Jul 19, 2026
7
i2_s ARM CPU kernel uses wrong weight-group layout (64/16 vs 128/32) → garbage output on all ARM CPU inference
microsoft/BitNet#585 · opened on Jul 15, 2026
2
deosnt compile and run out of the box
microsoft/BitNet#578 · opened on Jun 28, 2026
AVX-512 instructions enable denser packing and more efficient lookups.
microsoft/BitNet#569 · opened on Jun 14, 2026
10
Here are the current steps for builing BitNet.cpp for Raspberry Pi 5
microsoft/BitNet#561 · opened on May 20, 2026
1
Warning: huggingface-cli is deprecated and no longer works. Use hf instead.
microsoft/BitNet#560 · opened on May 20, 2026
2