Repository navigation
Repetition penalty #4
Description
Hello,
Thank you for this implementation, it is nice being able to experiment with things, even without GPUs at hand.
Would you mind implementing the repetition penalty? It seems to produce better/more consistent results...
Reacted by Peter, Oleksandr Nikitin, G2G2G2G and Ali Araabi
Activity
Merge pull request #4 from anon998/logging
add fallback for m chip & fix compiler bugs (ggml-org#4)
Add solver (ggml-org#4)
Merge pull request ggml-org#4 from jpgaribotti/QVAC-4552
Fix corrupted memory error on json grammar initialization
common : Fix corrupted memory error on json grammar initialization (#…
Merge pull request ggml-org#4 from Thireus/ug/port-sweep-bench
Merge pull request ggml-org#4 from ganminghao/graph
ggml webgpu: minor set rows optimization (#16810)
ggml webgpu: add support for emscripten builds (#17184)
138 remaining items
fix(sycl): exec-graph lifecycle -- wait-before-destroy + replay-futil…
feat(sycl): L1/L2 stats + DKW0 diagnostic hooks (reproducer instruments)
fix(sycl): second spec-review round for in-flight lease eviction test…
test(sycl): fix 0.0258 citation, scorer abs_floor + hard size-fail, F…
fix(sycl): submit the G4 numerics test's kernel on the unified-cache …
llama: LLAMA_GRAPH_TIMING, where the CPU time of a decode goes (ggml-…
Merge pull request ggml-org#4 from phase-4-cross-apu-refinement
strix-halo: fork overlay (docs + findings + model-search tooling)
common : Fix corrupted memory error on json grammar initialization (g…
ggml webgpu: minor set rows optimization (ggml-org#16810)
ggml webgpu: add support for emscripten builds (ggml-org#17184)
ggml webgpu: initial flashattention implementation (ggml-org#18610)
llama: LLAMA_GRAPH_TIMING, where the CPU time of a decode goes (ggml-…