v0.11.1

What's Changed

• other(auto): use RBLNAutoConfig.from_pretrained instead of removed .load by @rebel-junamsong in #612
• release: v0.11.0.post1 (rebuilt on v0.11.0 base) by @rebel-junamsong in #617
• other: revert misplaced post1 commit and back-merge main into dev by @rebel-junamsong in #618
• release: sync with main by @rebel-kblee in #609
• chore: advance main to dev (SHA parity, no code change) by @rebel-junamsong in #619
• other(deprecation): apply cutoff to pre-, post-, and dev-releases by @rebel-junamsong in #614
• model(seq2seq): raise instead of silently truncating over-long encoder input by @rebel-jongho in #620
• model: fix propagate rbln_config and add test by @rebel-seinpark in #538
• model: fix integrate seq idx to B,1 by @rebel-kblee in #622
• refactor(ops): drop duplicate custom op registration, defer to rebel-compiler by @rebel-jongho in #621
• other: Cleaner kv cache control by @rebel-jongho in #626
• dependency: Update dependency rebel-compiler to 0.11.1.dev501+g8b9056e9.prod by @rebel-thkim in #627
• other: resolve latest compiler version from Obedients nightly build by @rebel-thkim in #632
• other: manual update of rebel-compiler by @rebel-jongho in #633
• performance: add memory_budget for kv-cache block estimation by @rebel-jongho in #637
• model: memory_budget config semantics (non-save, mutually exclusive with kvcache_num_blocks) by @rebel-jongho in #639
• refactor: modernize type annotations to Python 3.10+ syntax by @rebel-jongho in #624
• other: modernize examples and README by @rebel-jongho in #635
• other: allow 1k kvcache_partition_len by @rebel-thkim in #642
• model: NPU-aware prefill_chunk_size default (512 on RBLN-CR) by @rebel-thkim in #634
• refactor: remove deprecations past their removal window for v0.11.1 by @rebel-thkim in #645
• model: fix default prefill_chunk_size to max_seq_len for bidirectional prefill (paligemma) by @rebel-thkim in #646
• model: restore batch_size floor in num_min_blocks by @rebel-jongho in #648
• model: Qwen3.5 by @rebel-seinpark in #631
• model(qwen3_5): fix video_grid_thw indexing for batched prefill by @rebel-seinpark in #651
• doc: reword the insufficient-kv-cache-blocks error by @rebel-jongho in #652
• model: precompute Qwen3.5 ViT rotary cos/sin tables once at load by @rebel-seinpark in #655
• model(decoderonly): return the full sequence when prompt length is a multiple of prefill_chunk_size by @rebel-thkim in #657
• model: blocking when gdn_chunk_size > 128 by @rebel-seinpark in #654
• model: gemma4 bug fix by @rebel-dkhong in #653
• model: revert Gemma4 rbln_config dtype casting by @rebel-dkhong in #660
• release: v0.11.1 by @rebel-thkim in #661

Full Changelog: v0.11.0...v0.11.1