Nexa SDK Files

Hexagon HTP

Microsoft-signed HTP catalog (llama.cpp @ 873e5d). No cert import required on Windows on Snapdragon.

What's Changed

• ci(bench): add gemma-4 GPU cells and a --compute filter to QDC bench by @mengshengwu in https://github.com/qualcomm/GenieX/pull/1335
• fix(server): report speculative-decoding stats in OpenAI usage by @mengshengwu in https://github.com/qualcomm/GenieX/pull/1336
• ci(bench): add gemma-4-26B-A4B-it QAT GGUF for QDC bench by @mengshengwu in https://github.com/qualcomm/GenieX/pull/1347
• fix(bench): run VLM llama_cpp cells as text LLM in ctx/pp/tg sweep by @mengshengwu in https://github.com/qualcomm/GenieX/pull/1343
• fix(sdk): default llama.cpp swa_full to false by @mengshengwu in https://github.com/qualcomm/GenieX/pull/1346
• feat(server): parse Gemma 4 tool-call format by @RemiliaForever in https://github.com/qualcomm/GenieX/pull/1345
• feat(sdk): add MTP speculative-decoding coverage to bench matrix by @mengshengwu in https://github.com/qualcomm/GenieX/pull/1348
• feat(sdk): return PROMPT_TOO_LONG when the prompt itself overflows by @RemiliaForever in https://github.com/qualcomm/GenieX/pull/1349
• feat(sdk): measure media encoder time separately from prefill by @RemiliaForever in https://github.com/qualcomm/GenieX/pull/1354
• fix(cli): drop unsupported modalities before VLM generate by @RemiliaForever in https://github.com/qualcomm/GenieX/pull/1356
• fix(cli): run update installer silently by @mansiverma897993 in https://github.com/qualcomm/GenieX/pull/1352
• test(sdk): add CPU + qairt unit tests for chat template, KV shift, mtmd by @mengshengwu in https://github.com/qualcomm/GenieX/pull/1355
• fix(bench): raise QDC job timeout and drop hybrid from the matrix by @mengshengwu in https://github.com/qualcomm/GenieX/pull/1358
• test(sdk): add cross-backend logits parity check by @mengshengwu in https://github.com/qualcomm/GenieX/pull/1357
• test(qdc): skip test_mtp_multi_turn on QCS9075M by @mengshengwu in https://github.com/qualcomm/GenieX/pull/1359
• feat(sdk): inject tools into the QAIRT VLM chat template by @RemiliaForever in https://github.com/qualcomm/GenieX/pull/1360
• feat(sdk): accept an explicit device list as a compute unit by @mengshengwu in https://github.com/qualcomm/GenieX/pull/1361
• fix(sdk): reuse VLM KV via char-level prefix match by @RemiliaForever in https://github.com/qualcomm/GenieX/pull/1363
• test(python): disable thinking in llm quality keyword cells by @RemiliaForever in https://github.com/qualcomm/GenieX/pull/1365
• fix(resume): re-fetch file when size mismatches legacy no-marker state by @mengshengwu in https://github.com/qualcomm/GenieX/pull/1364
• feat(sdk): add a CPU-only linux build variant for baseline armv8.0 boards by @Davidqian123 in https://github.com/qualcomm/GenieX/pull/1340
• feat(sdk): resolve AIHM version from latest.txt instead of pinning by @Hahahaooovvv in https://github.com/qualcomm/GenieX/pull/1353
• test(sdk): stop leaking tempdirs in model-manager store tests by @RemiliaForever in https://github.com/qualcomm/GenieX/pull/1368
• feat(android): add a CPU-only AAR for baseline armv8.0 devices by @RemiliaForever in https://github.com/qualcomm/GenieX/pull/1366
• docs(docs): document multi-HTP device list in MTP tutorial and CLI reference by @mengshengwu in https://github.com/qualcomm/GenieX/pull/1370
• feat(sdk): fetch AIHM index from releases/latest/, pinned as fallback by @Hahahaooovvv in https://github.com/qualcomm/GenieX/pull/1372
• chore: bump up llama.cpp by @vinovo in https://github.com/qualcomm/GenieX/pull/1374

Full Changelog: https://github.com/qualcomm/GenieX/compare/v0.4.0...v0.5.0