v0.7.0 — H4 hardfork (new models, PoM proof v2) + telemetry
v0.7.0 — AMD: H4 hardfork (new models, PoM proof v2) + miner telemetry
This is the AMD build of the H4 hardfork release. It is H4-ready and keeps mining the current lineup until the fork activates, then auto-switches with no restart.
H4 hardfork (activates at DAA 54,766,000)
- New model lineup (5 tiers): EXAONE-4.0-1.2B (light), Mistral-7B-v0.3, GLM-4-9B-0414, Qwen3.6-27B, Kimi-Linear-48B — all served by the bundled llama.cpp engine.
- PoM proof v2 (coin-age verification): at/after the gate the miner emits a recompute-from-chunks proof (every walked chunk Merkle-proven under the tier root) that the H4 node verifier accepts. Selected automatically by block height — no configuration.
- Era-aware: pre-fork blocks keep the old lineup + proof v1, so mining stays valid across the crossing; the switch is automatic at the gate.
- Staging override for testing:
KERYX_H4_ACTIVATION_DAA.
Miner telemetry
Optional mining.hello / mining.telemetry stratum messages (rig identity + per-GPU stats) for pool dashboards. Best-effort and non-fatal — never blocks or affects share submission; hostname is opt-in. Turn off entirely with KERYX_TELEMETRY_DISABLE=1.
AMD specifics
- The PoM walk (OpenCL) and the zero-dup engine are unchanged and verified byte-exact against the pinned tier root and a fixed winning nonce on the merged tree.
- The bundled llama.cpp already supports all five H4 model architectures — no separate engine update needed.
- Also carries the v0.6.11.2 discrete-GPU selection fix (engine never lands on an integrated GPU).
AMD assets
| file | sha256 |
|---|---|
keryx-miner-supr-amd-0.7.0.0.tar.gz (HiveOS) |
22746dd812f99eb4a49d270017cc717e2fbfd4f19368d5451f7c65a39c73a304 |
keryx-miner-supr-amd-mmpos_0.7.0.0.tar.gz (mmpOS) |
055398d9f9ffb3b47da141343fa7e3c29ed7411590c37e38481fa21dce7225bf |
keryx-miner-supr-amd-0.7.0.0.zip (Linux runnable) |
ac92b6e268707916c526eb1b69ef67d4cf6ea179f05ea126e12a7dd546bae190 |
Windows AMD build follows via CI.