0.16.0

List view

[Feature]: the ram usage of quantizing gemma-4-26B-A4B-it is too high

enhancementNew feature or requestNew feature or request

Status: Open.

#2210 In intel/auto-round;

· wenhuach21 opened on Aug 21, 2026

Enhance AutoScheme

Status: Open.

Task

#910 In intel/auto-round;

· wenhuach21 opened on Oct 17, 2025

Training-free sparse attention explore

high priority

Status: Open.

#2024 In intel/auto-round;

· chensuyue opened on Jul 8, 2026

[Feature]: each calibration dataset set its own default apply_chat_template value

enhancementNew feature or requestNew feature or request

Status: Open.

#2099 In intel/auto-round;

· wenhuach21 opened on Jul 29, 2026

Qwen3-30B-A3B-Instruct-2507 bf16 vllm eval failed on H200

Status: Open.

#2151 In intel/auto-round;

· chensuyue opened on Aug 12, 2026

[discussions]: [Compatibility/vLLM] AutoScheme mixed-bit export (llm_compressor) fails in vLLM with AssertionError in load_merged_column_weight

bugSomething isn't workingSomething isn't working

Status: Open.

#2174 In intel/auto-round;

· vishvaRam opened on Aug 17, 2026

[Bug]: woq_gemm's m > 1 path submits its oneDNN matmul to a cached stream bound to the first queue ever seen, so it is never recorded into a torch.xpu graph and every replay returns zeros

Status: Open.

#2206 In intel/auto-round;

· alankessler opened on Aug 21, 2026

[UT] tiny_qwen_moe_model_path loads the full Qwen1.5-MoE-A2.7B checkpoint

Status: Open.

Bug

#2267 In intel/auto-round;

· XuehaoSun opened on Sep 1, 2026

[Bug]: Fix AutoRound gradients for Kimi-VL routed experts- #2097

bugSomething isn't workingSomething isn't working

Status: Open.

#2282 In intel/auto-round;

· wenhuach21 opened on Sep 3, 2026

Stream targeted block re-quantization's calibration pass

Status: Open (in progress).

intel/auto-roundnumber 2064

#2064 In intel/auto-round;

· aquilarubra opened on Jul 19, 2026

Fix AutoRound gradients for Kimi-VL routed experts

Status: Draft (not ready).

intel/auto-roundnumber 2097

#2097 In intel/auto-round;

· lesj0610 opened on Jul 28, 2026

Quality Assurance for v0.16.0

high priority

Status: Open.

#2283 In intel/auto-round;

· XuehaoSun opened on Sep 3, 2026

[Upstream]: vLLM inference with AR kernels

high priority

Status: Open.

#1446 In intel/auto-round;

· luoyu-intel opened on Feb 12, 2026

[LLMC] llama3.1_W8A8_FP8_Block: ValueError: AutoRoundModifier only supports channel-wise and tensor-wise weight quantization

Status: Open.

Bug

#2196 In intel/auto-round;

· XuehaoSun opened on Aug 20, 2026

[LLMC] llama3_wNa16_FP8KV: AttributeError: 'NoneType' object has no attribute 'group_size'

Status: Open.

Bug

#2195 In intel/auto-round;

· XuehaoSun opened on Aug 20, 2026

[Bug]: MTP logic is unaware of Transformers' layer renaming and quantizes the entire model again

bugSomething isn't workingSomething isn't working

Status: Open.

#2324 In intel/auto-round;

· wenhuach21 opened on Sep 9, 2026

feat: data-parallel block tuning via --parallel_quantization

Status: Draft (not ready).

intel/auto-roundnumber 2351

#2351 In intel/auto-round;

· avtc opened on Sep 11, 2026

[Bug]: MiniMax-M2.7_mxfp4_mixed_fp8kv_LLMC vllm eval failed

bugSomething isn't workingSomething isn't working

Status: Open.

Bug

#2369 In intel/auto-round;

· chensuyue opened on Sep 16, 2026

[Bug]: Kimi-K2.6_mxfp4_fp8attn_LLMC vllm eval qkv scale is not loaded

bugSomething isn't workingSomething isn't working

Status: Open.

Bug

#2370 In intel/auto-round;

· chensuyue opened on Sep 16, 2026

Don't rewrite packed modules as float weights when resuming

Status: Open (in progress).

intel/auto-roundnumber 2386

#2386 In intel/auto-round;

· Shubham-Padkonde opened on Sep 20, 2026

[Qwen3-8B-FP8] RuntimeError: Trying to backward through the graph a second time

Status: Open.

Bug

#2387 In intel/auto-round;

· XuehaoSun opened on Sep 20, 2026

fix: update get_predefined_ignore_layers to include shared_expert_gat…

Status: Open (in progress).

intel/auto-roundnumber 2397

#2397 In intel/auto-round;

· xin3he opened on Sep 21, 2026

[Bug]: accuracy regression of v2

bugSomething isn't workingSomething isn't working

high priority

Status: Open.

#2400 In intel/auto-round;

· wenhuach21 opened on Sep 21, 2026

revert pr 2360

bugSomething isn't workingSomething isn't working

Status: Open.

#2403 In intel/auto-round;

· wenhuach21 opened on Sep 21, 2026