Repository navigation

Add structured ranking and decision runtime APIs - #2649

Merged

Akshay Sonawane (apsonawane) merged 38 commits into

mainfrom

asonawane/non-generative-runtime

Oct 29, 2026

Merged

Add structured ranking and decision runtime APIs#2649
Akshay Sonawane (apsonawane) merged 38 commits into
mainfrom
asonawane/non-generative-runtime

Commits on Oct 28, 2026
- Add structured non-generative runtime APIs
  Akshay Sonawane (apsonawane)
  committed
  626a519
- Test and document non-generative runtime APIs
  Akshay Sonawane (apsonawane)
  committed
  1386229
- Harden non-generative runtime boundaries
  Akshay Sonawane (apsonawane)
  committed
  d33c20b
- Fix non-generative runtime portability
  Akshay Sonawane (apsonawane)
  committed
  f24a35c
- Fix lint
  Akshay Sonawane (apsonawane)
  committed
  9c2a41c
- Fix runtime portability and empty tensor handling
  Akshay Sonawane (apsonawane)
  committed
  22066a5
- Fix Windows runtime warnings
  Akshay Sonawane (apsonawane)
  committed
  a5cfb51
- Close runtime coverage and numeric gaps
  Akshay Sonawane (apsonawane)
  committed
  24f926c
- Honor exported KEV tensor contracts
  Akshay Sonawane (apsonawane)
  committed
  7ef5a70
- Contain legacy component paths
  Akshay Sonawane (apsonawane)
  committed
  70ce27d
- Optimize short KEV CUDA prefixes
  Akshay Sonawane (apsonawane)
  committed
  8c2d6d8
- Add opt-in KEV CUDA graph capture
  Akshay Sonawane (apsonawane)
  committed
  97b8276
- Cache CLM state projections
  Akshay Sonawane (apsonawane)
  committed
  b08db18
- Load component session options from config
  
  Akshay Sonawane (apsonawane)
  and
  Copilot
  committed
  0157d97
- Accelerate fused CLM ranking with CUDA graph buckets
  Akshay Sonawane (apsonawane)
  committed
  eaddfce
- Address typed runtime qualification review
  
  Akshay Sonawane (apsonawane)
  and
  Copilot
  committed
  e85990e
- Reuse existing tensor and tokenizer APIs
  
  Akshay Sonawane (apsonawane)
  and
  Copilot
  committed
  9406b42
- Address cross-platform typed runtime review
  
  Akshay Sonawane (apsonawane)
  and
  Copilot
  committed
  c753aef
- Minimize typed runtime public API
  
  Akshay Sonawane (apsonawane)
  and
  Copilot
  committed
  fe4bd8e
- Keep typed runtime API production-focused
  
  Akshay Sonawane (apsonawane)
  and
  Copilot
  committed
  1b2002f
- Collapse typed runtime to JSON workflows
  
  Akshay Sonawane (apsonawane)
  and
  Copilot
  committed
  094658d
- Remove unused structured CLM workflow
  
  Akshay Sonawane (apsonawane)
  and
  Copilot
  committed
  f9e662f
- Unify non-generative session API
  
  Akshay Sonawane (apsonawane)
  and
  Copilot
  committed
  bbdb2a7
- Remove duplicate session ownership state
  
  Akshay Sonawane (apsonawane)
  and
  Copilot
  committed
  df74944
- Reuse OgaConfig for non-generative sessions
  Akshay Sonawane (apsonawane)
  committed
  57c3f1d
- Remove unused dimension override wrapper
  Akshay Sonawane (apsonawane)
  committed
  1757af0
- Inline non-generative string ownership
  Akshay Sonawane (apsonawane)
  committed
  4d4cecb
- Centralize non-generative C API errors
  Akshay Sonawane (apsonawane)
  committed
  9f5eb5d
- Use model config for typed tokenization
  Akshay Sonawane (apsonawane)
  committed
  cea21c6
- Simplify typed session cache keys
  Akshay Sonawane (apsonawane)
  committed
  d68505b
- Remove obsolete fixture changes
  Akshay Sonawane (apsonawane)
  committed
  caa279e
- Remove redundant C++ visibility macro
  Akshay Sonawane (apsonawane)
  committed
  2150afb
- Reuse Config in Python typed sessions
  Akshay Sonawane (apsonawane)
  committed
  613241a
- Simplify non-generative Python binding
  Akshay Sonawane (apsonawane)
  committed
  38aa7ed
- Address non-generative runtime review feedback
  Akshay Sonawane (apsonawane)
  committed
  aff4965
- Separate Unicode and path option tests
  Akshay Sonawane (apsonawane)
  committed
  4c2fe63
- Load package tokenizer data from memory
  Akshay Sonawane (apsonawane)
  committed
  5e3b948
- Merge main into non-generative runtime
  Akshay Sonawane (apsonawane)
  committed
  7ba4c90

Commits

Commits on Oct 28, 2026