chore(main): release 1.3.0 (#48)

🤖 I have created a release *beep* *boop*
---


##
[1.3.0](v1.2.2...v1.3.0)
(2026-03-31)


### Features

* **kv-cache:** add sliding window attention bypass for Gemma models
([#53](#53))
([ee5300e](ee5300e))
* **triton:** add head_dim 64/96 kernel support with non-pow2 padding
([#52](#52))
([44edd60](44edd60))
* **verify:** add head_dim 256 support and validate Gemma-2/Gemma-3
([#55](#55))
([e0f5d45](e0f5d45))
* **verify:** validate Phi-3-mini compression quality
([#54](#54))
([f5fec46](f5fec46))
* **verify:** validate Qwen2.5-3B and Phi-4 compression quality
([#51](#51))
([48243c9](48243c9))


### Bug Fixes

* **benchmark:** add head_dim detection division guard
([#56](#56))
([380a7eb](380a7eb))
* **triton:** add defensive assertions and throughput penalty docs
([44edd60](44edd60))
* **vllm:** guard forward() against kv_cache=None during profiling
([#57](#57))
([3c5fdde](3c5fdde))


### Performance Improvements

* **experiments:** add experiment 022 clip duration comparison
([#47](#47))
([55503a9](55503a9))
* **experiments:** add experiment 023 frame count sweep
([#49](#49))
([d025208](d025208))
* **experiments:** add experiment 024 zero-change model probe
([#50](#50))
([d11fc3a](d11fc3a))


### Documentation

* **roadmap:** add experiment 023 frame count sweep findings
([d025208](d025208))

---
This PR was generated with [Release
Please](https://github.com/googleapis/release-please). See
[documentation](https://github.com/googleapis/release-please#release-please).

---------

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

Author: Alberto-Codes

.release-please-manifest.json

@@ -1,3 +1,3 @@
 {
-  ".": "1.2.2"
+  ".": "1.3.0"
 }

CHANGELOG.md

@@ -1,5 +1,35 @@
 # Changelog
 
+## [1.3.0](https://github.com/Alberto-Codes/turboquant-vllm/compare/v1.2.2...v1.3.0) (2026-03-31)
+
+
+### Features
+
+* **kv-cache:** add sliding window attention bypass for Gemma models ([#53](https://github.com/Alberto-Codes/turboquant-vllm/issues/53)) ([ee5300e](https://github.com/Alberto-Codes/turboquant-vllm/commit/ee5300e30954d69ed2c0efa5d41ee9c2821a0441))
+* **triton:** add head_dim 64/96 kernel support with non-pow2 padding ([#52](https://github.com/Alberto-Codes/turboquant-vllm/issues/52)) ([44edd60](https://github.com/Alberto-Codes/turboquant-vllm/commit/44edd603f481d9dfc99af6173211bc3e034bc7c1))
+* **verify:** add head_dim 256 support and validate Gemma-2/Gemma-3 ([#55](https://github.com/Alberto-Codes/turboquant-vllm/issues/55)) ([e0f5d45](https://github.com/Alberto-Codes/turboquant-vllm/commit/e0f5d45b63411f7f7fa83f8c2a55b2382a35e922))
+* **verify:** validate Phi-3-mini compression quality ([#54](https://github.com/Alberto-Codes/turboquant-vllm/issues/54)) ([f5fec46](https://github.com/Alberto-Codes/turboquant-vllm/commit/f5fec464ecf49d314cbf879e745ee81b9d7dc886))
+* **verify:** validate Qwen2.5-3B and Phi-4 compression quality ([#51](https://github.com/Alberto-Codes/turboquant-vllm/issues/51)) ([48243c9](https://github.com/Alberto-Codes/turboquant-vllm/commit/48243c97146f7f712e07531cd82ba844fe031721))
+
+
+### Bug Fixes
+
+* **benchmark:** add head_dim detection division guard ([#56](https://github.com/Alberto-Codes/turboquant-vllm/issues/56)) ([380a7eb](https://github.com/Alberto-Codes/turboquant-vllm/commit/380a7eb2554f5f81d419a7438967219df0db05ae))
+* **triton:** add defensive assertions and throughput penalty docs ([44edd60](https://github.com/Alberto-Codes/turboquant-vllm/commit/44edd603f481d9dfc99af6173211bc3e034bc7c1))
+* **vllm:** guard forward() against kv_cache=None during profiling ([#57](https://github.com/Alberto-Codes/turboquant-vllm/issues/57)) ([3c5fdde](https://github.com/Alberto-Codes/turboquant-vllm/commit/3c5fdde4d139662c389c62e2a2664a0b58a246a9))
+
+
+### Performance Improvements
+
+* **experiments:** add experiment 022 clip duration comparison ([#47](https://github.com/Alberto-Codes/turboquant-vllm/issues/47)) ([55503a9](https://github.com/Alberto-Codes/turboquant-vllm/commit/55503a9b7e48738e0c38e88264ee2361e376899e))
+* **experiments:** add experiment 023 frame count sweep ([#49](https://github.com/Alberto-Codes/turboquant-vllm/issues/49)) ([d025208](https://github.com/Alberto-Codes/turboquant-vllm/commit/d0252089cd8cd75aff9019fea82b19e74d3b6be1))
+* **experiments:** add experiment 024 zero-change model probe ([#50](https://github.com/Alberto-Codes/turboquant-vllm/issues/50)) ([d11fc3a](https://github.com/Alberto-Codes/turboquant-vllm/commit/d11fc3a88cc97dd08e5b5027a9f7c5d5db8c8116))
+
+
+### Documentation
+
+* **roadmap:** add experiment 023 frame count sweep findings ([d025208](https://github.com/Alberto-Codes/turboquant-vllm/commit/d0252089cd8cd75aff9019fea82b19e74d3b6be1))
+
 ## [1.2.2](https://github.com/Alberto-Codes/turboquant-vllm/compare/v1.2.1...v1.2.2) (2026-03-30)
 
 

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "turboquant-vllm"
-version = "1.2.2"
+version = "1.3.0"
 description = "TurboQuant KV cache compression for vLLM — fused Triton kernels, 3.76x compression, 3.7x faster decode on RTX 4090"
 readme = "README.md"
 license = { text = "Apache-2.0" }