-
-
Notifications
You must be signed in to change notification settings - Fork 210
Pull requests: dphnAI/sonar
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
fix(exl3): clamp out-of-range expert ids in MoE prefill scatter path
#1782
opened Sep 9, 2026 by
Nakanokensetsu
Contributor
Loading…
fix(exl3): give the MoE down projection its own Hadamard scratch
#1775
opened Sep 6, 2026 by
Andybui1012
Loading…
feat: resolve oci:// model references via llmman serve
#1769
opened Aug 30, 2026 by
ericcurtin
Loading…
2 of 4 tasks
fix: thinking_token_budget has no effect when async scheduling is enabled
#1766
opened Aug 13, 2026 by
Nakanokensetsu
Contributor
Loading…
3 tasks done
Don't disable other samplers for non-mirostat rows in a mixed batch
#1674
opened Jun 5, 2026 by
lollinng
Loading…
[kernel][mobile] add mobile-optimized CPU backend
#1609
opened Nov 15, 2025 by
AlpinDale
Member
Loading…
[Build] feat: multi-backend build system to consolidate CUDA, ROCm, CPU, etc
#1540
opened Sep 29, 2025 by
AlpinDale
Member
Loading…
feat(compilation): Extend sequence parallelism to activations
#1523
opened Sep 18, 2025 by
AlpinDale
Member
Loading…
[Kernel][Quantization] feat: add Gluon kernels for AWQ quantization
#1520
opened Sep 17, 2025 by
AlpinDale
Member
Loading…
fix: illegal memory access for FP8 MoE models in cutlass 3x grouped gemm kernel
#1446
opened Aug 28, 2025 by
AlpinDale
Member
Loading…
[Kernel] feat: add custom CUDA kernels for all sampling ops
#1444
opened Aug 26, 2025 by
AlpinDale
Member
Loading…
gguf: optimize prefill speeds for Q4_K quants
#1395
opened Jul 18, 2025 by
AlpinDale
Member
Loading…
Fix metrics and allow disable block manager v2
#1299
opened Apr 15, 2025 by
Nero10578
Contributor
Loading…
Previous Next
ProTip!
Follow long discussions with comments:>50.