Skip to content

fix(cpu): remove hardcoded paths and unused arm_neon include#477

Merged
chenghuaWang merged 5 commits intoUbiquitousLearning:v2from
chenghuaWang:v2
Oct 14, 2025
Merged

fix(cpu): remove hardcoded paths and unused arm_neon include#477
chenghuaWang merged 5 commits intoUbiquitousLearning:v2from
chenghuaWang:v2

Conversation

@chenghuaWang
Copy link
Copy Markdown
Collaborator

  • Replace absolute paths with relative includes for x86 kernel headers
  • Remove unnecessary arm_neon.h include in radix attention implementation

…ment

- Added `kvCache()` method to expose KV cache in `Qwen3ForCausalLM`
- Adjusted `makeRotaryPosEmbedding` function signature for better formatting
- Added new playground example for cache-reduced computation benchmarking
- Included CMake and Python build scripts for the new example
- Updated .gitignore to exclude SDK directories in playground
- Replace absolute paths with relative includes for x86 kernel headers
- Remove unnecessary arm_neon.h include in radix attention implementation
- Expose `kvCache()` method in `Qwen2VLForCausalLM` to allow external access
  to the key-value cache for reduced computation scenarios
- Add `qwen2vlLlmPartBench` function to benchmark Qwen2VL LLM part with
  configurable prefix cache length and prefill tokens
- Update main function to support model selection between qwen2vl-2B and
  qwen3-0.6B for running corresponding benchmarks
- Include necessary headers for Qwen2VL model components
- Add python script for compiling and deploying the benchmark to Android
  using ADB toolkit
@chenghuaWang chenghuaWang merged commit aaaedf5 into UbiquitousLearning:v2 Oct 14, 2025
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant