|
|
@@ -111,8 +111,12 @@ anchors above). Tasked for the next window:
|
|
|
model `deepseek-ai/DeepSeek-V4-Flash(-Vision)` rev-pinned, **SPARK_KV_CACHE
|
|
|
= 26214034944 B ≈ 24.4 GiB @ GMU 0.82** (independent confirmation of the
|
|
|
24 GiB/rank headroom theory), CUDAGRAPH_MODE switch (FULL_AND_PIECEWISE).
|
|
|
- Where "vision" fits: repo tag 3.72-vision (2026-09-03) was already pulled
|
|
|
- on all nodes; 3.73-vision is today's follow-up.
|
|
|
+ Where "vision" fits (operator confirmation): **this is the
|
|
|
+ DeepSeek-V4-Flash-**Vision** aiden lane** — a direct competitor to our
|
|
|
+ fork lane `dsv4-flash-vision` (DSpark-6) and the eugr image that
|
|
|
+ FAILED to carry V4-Vision at all. The 'vision' branch effectively
|
|
|
+ replaces the blocked dsv4-flash-vision-eugr row with an actionable one.
|
|
|
+ Existing recipes `dsv4-vision-aiden-2x/4x` point at this family already.
|
|
|
|
|
|
Status ledger (main matrix above): DONE = r17abi baseline, r18 2x/4x/DCP4/
|
|
|
nvfp4KV, RedHat MTP, eugr 2x(FAIL), spark-2x-nvfp4kv (158, best 2x); BLOCKED =
|
