Repository navigation
Conversation
…ime gates, CUDART constant shim, calibrated hipBLASLt table, docs and bench evidence
…oring the RDNA4 one), both models benched, Q2_0 at its full 256K context
Contributor
Author
|
Getting about 900/s on pp and 50/s on tg. System is a 7800 card in a zen4 ddr4 system. Only had a chance to do some light testing then rebased as the gfx1200 stuff came in, now getting back to testing. Testing is somewhat limited but results so far are quite useable. Going to take it on some fill to 256K and compress cycles to see how it fares. |
Niko1221
added a commit
that referenced
this pull request
Oct 1, 2026
…256 by Efeisot) as community-validated cards Cut down from #254 and #256 onto 0.1.30's AMD tables: - setup.py: AMD_ARCHS / AMD_NAMES / ROCM_INDEXES / AMD_CARDS list gfx1101 (TheRock gfx110X-dgpu) and gfx1200 (gfx120X-all); amd_gpus() takes the sysfs root so the detection can be tested. - cmake/hip_backend.cmake: gfx1101 and gfx1200 build without the "not validated" warning (a status line names them community-validated); gfx1102 (#192: ctest only) keeps the warning. - docs/AMD_HIP.md (Community-validated cards: the owners' setups, ctest and numbers) and README. - tools/test_setup_amd.py: detection on a mocked KFD topology (every family, the CPU node, an iGPU and gfx1102 listed as not supported, product-name fallbacks, the ROCm index per family). Left out: the PRs' own cmake/device.cu arch checks (0.1.30's compiled_for covers them), the dp4a change (main has it), math_constants.h, the bench JSONs, the launcher, and both hipBLASLt tables (neither PR measured the prompt speed without its table). Co-Authored-By: Joseph Hohertz <jhohertz@gmail.com> Co-Authored-By: efeisot <efeproofkmr@gmail.com> Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0154bwiad8gY7wRehCXy27pN
Owner
|
Released in 0.1.31: gfx1101 is a community-validated card in setup and the docs, with your report (without the bench JSONs and |
Niko1221
added a commit
that referenced
this pull request
Oct 6, 2026
…256 by Efeisot) as community-validated cards Cut down from #254 and #256 onto 0.1.30's AMD tables: - setup.py: AMD_ARCHS / AMD_NAMES / ROCM_INDEXES / AMD_CARDS list gfx1101 (TheRock gfx110X-dgpu) and gfx1200 (gfx120X-all); amd_gpus() takes the sysfs root so the detection can be tested. - cmake/hip_backend.cmake: gfx1101 and gfx1200 build without the "not validated" warning (a status line names them community-validated); gfx1102 (#192: ctest only) keeps the warning. - docs/AMD_HIP.md (Community-validated cards: the owners' setups, ctest and numbers) and README. - tools/test_setup_amd.py: detection on a mocked KFD topology (every family, the CPU node, an iGPU and gfx1102 listed as not supported, product-name fallbacks, the ROCm index per family). Left out: the PRs' own cmake/device.cu arch checks (0.1.30's compiled_for covers them), the dp4a change (main has it), math_constants.h, the bench JSONs, the launcher, and both hipBLASLt tables (neither PR measured the prompt speed without its table). Co-Authored-By: Joseph Hohertz <jhohertz@gmail.com> Co-Authored-By: efeisot <efeproofkmr@gmail.com>
anon761
pushed a commit
to anon761/Strata-upstream
that referenced
this pull request
Oct 6, 2026
…060 XT, Niko1221#256 by Efeisot) as community-validated cards Cut down from Niko1221#254 and Niko1221#256 onto 0.1.30's AMD tables: - setup.py: AMD_ARCHS / AMD_NAMES / ROCM_INDEXES / AMD_CARDS list gfx1101 (TheRock gfx110X-dgpu) and gfx1200 (gfx120X-all); amd_gpus() takes the sysfs root so the detection can be tested. - cmake/hip_backend.cmake: gfx1101 and gfx1200 build without the "not validated" warning (a status line names them community-validated); gfx1102 (Niko1221#192: ctest only) keeps the warning. - docs/AMD_HIP.md (Community-validated cards: the owners' setups, ctest and numbers) and README. - tools/test_setup_amd.py: detection on a mocked KFD topology (every family, the CPU node, an iGPU and gfx1102 listed as not supported, product-name fallbacks, the ROCm index per family). Left out: the PRs' own cmake/device.cu arch checks (0.1.30's compiled_for covers them), the dp4a change (main has it), math_constants.h, the bench JSONs, the launcher, and both hipBLASLt tables (neither PR measured the prompt speed without its table). Co-Authored-By: Joseph Hohertz <jhohertz@gmail.com> Co-Authored-By: efeisot <efeproofkmr@gmail.com>
anon761
pushed a commit
to anon761/Strata-upstream
that referenced
this pull request
Oct 6, 2026
…060 XT, Niko1221#256 by Efeisot) as community-validated cards Cut down from Niko1221#254 and Niko1221#256 onto 0.1.30's AMD tables: - setup.py: AMD_ARCHS / AMD_NAMES / ROCM_INDEXES / AMD_CARDS list gfx1101 (TheRock gfx110X-dgpu) and gfx1200 (gfx120X-all); amd_gpus() takes the sysfs root so the detection can be tested. - cmake/hip_backend.cmake: gfx1101 and gfx1200 build without the "not validated" warning (a status line names them community-validated); gfx1102 (Niko1221#192: ctest only) keeps the warning. - docs/AMD_HIP.md (Community-validated cards: the owners' setups, ctest and numbers) and README. - tools/test_setup_amd.py: detection on a mocked KFD topology (every family, the CPU node, an iGPU and gfx1102 listed as not supported, product-name fallbacks, the ROCm index per family). Left out: the PRs' own cmake/device.cu arch checks (0.1.30's compiled_for covers them), the dp4a change (main has it), math_constants.h, the bench JSONs, the launcher, and both hipBLASLt tables (neither PR measured the prompt speed without its table). Co-Authored-By: Joseph Hohertz <jhohertz@gmail.com> Co-Authored-By: efeisot <efeproofkmr@gmail.com> Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0154bwiad8gY7wRehCXy27pN
ScrambledEggsOnToast
pushed a commit
to ScrambledEggsOnToast/Strata
that referenced
this pull request
Oct 18, 2026
… advertising a refused family Niko1221#251: serve/server.py parses the engine's effective --short-read out of its own argv (last flag wins, a negative clamps to 0, atoll stops at the first non-digit) and the progress gate uses it instead of a hardcoded 64-token tail, so a configured 640 drops the right PP line and a configured 16 no longer drops a legitimate one. Two tests cover the gate and the parser. docs/DETAILS.md states the rule. Niko1221#252: a failed optional capture is a durability problem, not a serving problem. A checkpoint, parked snapshot or shared-prefix capture that fails on a copy, allocation, validation or publication is reported with one bounded stderr line and dropped; only a failed device sync (a GPU hang, a failed stage sync) still ends the request. At shutdown a capture failure is reported but no longer latches the store unhealthy, which used to skip finalization, leave the run-intent armed and make the next start retire every record of the execution domain. The policy and its log line live in program/checkpoint_preservation.hpp with a CPU-only test. The extracted read_part harness carries the ambient flag. Niko1221#250: the all-pinned cache refusal is the documented contract. Pins and pending publications are never spent, so reserve_snapshot/put/make_room refuse atomically and evict_oldest returns false; the carrier skips parking and the request completes. The contract is stated at reserve_snapshot, in DETAILS.md, and in the two skip-parking lines, which now name how many entries are pinned. A test fills a cache with two pinned histories and asserts every path refuses without spending a pin or losing either history. Niko1221#254: this fork keeps refusing the qwen35moe family and now says so consistently. strata --families prints model_families() (qwen4exp only), so setup's existing family check refuses before the 22.9 GB download; the refusal and --help name the separate strata-q36 executable; model_family.hpp no longer claims the CUDA/HIP builds run the family; QWEN36/BATCHING/DETAILS and setup's comment match. The dead strata->strata_qwen35 link and STRATA_HAVE_QWEN35 are removed; strata-q36 still builds the engine path. Checks: strata 0.1.45 builds; conversation_cache_test 5199, conversation_store_test 6327, model_family_test, checkpoint_preservation_test and prefill_interleave_test pass; serve.test_server 403 tests OK; the binary prints qwen4exp and refuses a qwen35moe file with exit 2.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This just adds support for gfx1101 as well.