Commits

Branch selector

 main

User selector

All users

Datepicker

All time

Commit History

Commits on Aug 4, 2025

• Merge pull request #42 from mikepapadim/refactor-packages

  Show description for 85f1875
  mikepapadim

  authored
  85f1875

  Copy full SHA for 85f1875

• Expand and organize imports in `MistralChatFormat` for improved clarity and structure.

  mikepapadim

  committed
  fb7ab86

  Copy full SHA for fb7ab86

• Reorganize package structure and update imports to use `org.beehive.gpullama3` namespace for improved modularity and clarity.

  mikepapadim

  committed
  6cb7ad2

  Copy full SHA for 6cb7ad2

• Update README.md

  mikepapadim

  authored
  d60d9ea

  Copy full SHA for d60d9ea

• Update README.md

  mikepapadim

  authored
  1c11b5b

  Copy full SHA for 1c11b5b

• Merge pull request #38 from mikepapadim/phi-3

  Show description for be93182
  mikepapadim

  authored
  be93182

  Copy full SHA for be93182

• Reformat Javadoc in `ModelType` for improved readability and alignment; remove redundant line breaks.

  mikepapadim

  committed
  3cc7ee6

  Copy full SHA for 3cc7ee6

• Apply suggestion from @Copilot

  Show description for 2df92cc
  

  mikepapadim

  and

  Copilot

  authored
  2df92cc

  Copy full SHA for 2df92cc

• Reorder fields in `Qwen3ChatFormat`, remove redundant line breaks, and fix formatting for improved readability and consistency.

  mikepapadim

  committed
  c49558f

  Copy full SHA for c49558f

Commits on Aug 3, 2025

• Reformat Javadoc and parameter alignment in `InferenceEngine` for enhanced readability and consistency; remove redundant comments and unused legacy code.

  mikepapadim

  committed
  dfb3a4b

  Copy full SHA for dfb3a4b

• Fix redundant spacing and reformat comments in `Phi3TornadoVMLayerPlanner` for improved readability.

  mikepapadim

  committed
  1adfb8d

  Copy full SHA for 1adfb8d

• Refactor `Phi3TornadoVMLayerPlanner` by reverting `fusedGateUpSiLUDown` to individual FFN tasks, updating worker grid mappings, and removing the fused kernel from `TransformerComputeKernelsLayered`…

  Show description for 3ce5f36
  mikepapadim

  committed
  3ce5f36

  Copy full SHA for 3ce5f36

• Refactor `Phi3TornadoVMLayerPlanner` by replacing `fusedGateUpSiLUDownOptimized` with `fusedGateUpSiLUDown`, simplifying `TransformerComputeKernelsLayered` logic, and removing redundant memory oper…

  Show description for 9cbc5e5
  mikepapadim

  committed
  9cbc5e5

  Copy full SHA for 9cbc5e5

• Refactor `Phi3TornadoVMLayerPlanner` by consolidating FFN tasks into `fusedGateUpSiLUDownOptimized`, removing redundant operations, and updating worker grid mappings for improved efficiency and mai…

  Show description for 3e1887d
  mikepapadim

  committed
  3e1887d

  Copy full SHA for 3e1887d

• Refactor `Phi3TornadoVMLayerPlanner` by consolidating QKV splitting tasks into `splitQKV`, removing redundant `copy` operations, and updating worker grid mappings for improved efficiency and mainta…

  Show description for c5b53b2
  mikepapadim

  committed
  c5b53b2

  Copy full SHA for c5b53b2

• Refactor `Phi3TornadoVMLayerPlanner` by consolidating tasks into `splitGateUpAndSiLU`, removing redundant steps, and updating worker grid mappings for improved maintainability and efficiency.

  mikepapadim

  committed
  2331deb

  Copy full SHA for 2331deb

• Reformat `TransformerComputeKernelsLayered` by adjusting Javadoc annotations, reorganizing line breaks, and improving parameter alignment for enhanced readability and consistency.

  mikepapadim

  committed
  47406da

  Copy full SHA for 47406da

• Fix spacing issue in `Phi3TornadoVMLayerPlanner` loop declaration for improved code readability and consistency.

  mikepapadim

  committed
  e60f478

  Copy full SHA for e60f478

• Clean up `Phi3ChatFormat` by simplifying imports, reordering fields, removing commented-out code, and improving Javadoc formatting for better readability and maintainability.

  mikepapadim

  committed
  f1c51aa

  Copy full SHA for f1c51aa

• Refactor `ChatFormat` by reordering methods, simplifying imports, and improving Javadoc formatting for enhanced readability and maintainability.

  mikepapadim

  committed
  b2a518c

  Copy full SHA for b2a518c

• Clean up `InferenceCore` by removing commented-out code, redundant lines, and unnecessary variables to improve readability and maintainability.

  mikepapadim

  committed
  34b3c79

  Copy full SHA for 34b3c79

• Enhance `Phi3Tokenizer` decoding by adding UTF-8 byte token handling, updating `encode` method implementation, and integrating byte buffer logic for improved support of special tokens.

  mikepapadim

  committed
  9f6b99e

  Copy full SHA for 9f6b99e

• Reformat `TornadoVMMasterPlan` by re-organizing line breaks, updating parameter annotations, and consolidating documentation for improved readability and consistency.

  mikepapadim

  committed
  8ffc282

  Copy full SHA for 8ffc282

• Clean up `Phi3Tokenizer` by reorganizing constants, reformatting code, and removing unused fields for improved readability and maintainability.

  mikepapadim

  committed
  0e1c68c

  Copy full SHA for 0e1c68c

• Refactor `ModelLoader` by reformatting code, removing redundant imports, and reorganizing methods for better readability and maintainability.

  mikepapadim

  committed
  3bf1f39

  Copy full SHA for 3bf1f39

• Clean up `Phi3TornadoVMLayerPlanner` by removing commented-out code and redundant line breaks for better readability and maintainability.

  mikepapadim

  committed
  c3528a0

  Copy full SHA for c3528a0

Commits on Aug 1, 2025

• Refactor `generateTokensGPUPhi3` for improved max token validation and performance metrics tracking

  mikepapadim

  committed
  686472e

  Copy full SHA for 686472e

• Implement rope rotation for Phi3 in `TransformerComputeKernelsLayered` and update task reference in `Phi3TornadoVMLayerPlanner`

  mikepapadim

  committed
  4428362

  Copy full SHA for 4428362

• Fix worker grid configuration in `Phi3TornadoVMLayerPlanner` for correct local work settings

  mikepapadim

  committed
  50a0909

  Copy full SHA for 50a0909

• Add `Phi3TornadoVMLayerPlanner` to manage TornadoVM-based forward pass planning and GPU task scheduling for Phi3 model

  mikepapadim

  committed
  80632f0

  Copy full SHA for 80632f0

• [WIP] Implement `generateTokensGPUPhi3` and integrate `Phi3TornadoVMLayerPlanner` with GPU acceleration logic

  mikepapadim

  committed
  e35d4de

  Copy full SHA for e35d4de

• Add TornadoVM wrappers in `Phi3State` for GPU acceleration and clean up unused code in `InferenceCore`

  mikepapadim

  committed
  e54d13c

  Copy full SHA for e54d13c

• Update `Model` logic to correctly handle `PHI_3` model type in conditional check

  mikepapadim

  committed
  1ec8ace

  Copy full SHA for 1ec8ace

• Complete `generateTokensPhi3` implementation and refine Phi3 tokenizer logic

  mikepapadim

  committed
  8adb2bb

  Copy full SHA for 8adb2bb

• Integrate Phi3-specific state management, model loading, and tokenizer logic

  mikepapadim

  committed
  104a5bb

  Copy full SHA for 104a5bb

Pagination

Previous

Next