Upload docs/TEARDOWN_READY_l3long_2026-09-09.txt with huggingface_hub
Browse files
docs/TEARDOWN_READY_l3long_2026-09-09.txt
CHANGED
|
@@ -18,6 +18,20 @@ UPLOADED
|
|
| 18 |
checkpoints/l3long_final_run17_step_2450_rank{0,1}.ckpt
|
| 19 |
checkpoints/l3long_final_run17_step_2450_rank{0,1}.zero_g1 <- optimizer
|
| 20 |
state, so step 2450 is RESUMABLE
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
|
| 22 |
TO RESUME l3long LATER
|
| 23 |
Pull the four step_2450 files, restore them as
|
|
@@ -30,10 +44,8 @@ STILL NOT DONE
|
|
| 30 |
2. ext_32k.sh (parent l2, b3 data) -- written, never run. NOTE a DIFFERENT
|
| 31 |
32k extension (ext32k_b4, b4 data) WAS run and evaluated earlier today;
|
| 32 |
see docs/EXT32K_RESULTS_2026-09-09.md and logs/ext32k/.
|
| 33 |
-
3.
|
| 34 |
-
|
| 35 |
-
tarballs above are the only backup -- commit them yourself before relying
|
| 36 |
-
on git history.
|
| 37 |
|
| 38 |
NOT UPLOADED (deliberate, re-derivable)
|
| 39 |
lendata/ 3.3 GB regenerate with fetch_eval_assets.sh
|
|
|
|
| 18 |
checkpoints/l3long_final_run17_step_2450_rank{0,1}.ckpt
|
| 19 |
checkpoints/l3long_final_run17_step_2450_rank{0,1}.zero_g1 <- optimizer
|
| 20 |
state, so step 2450 is RESUMABLE
|
| 21 |
+
checkpoints/l3long1000_run17_step_1000_rank{0,1}.zero_g1
|
| 22 |
+
checkpoints/l3long2000_run17_step_2000_rank{0,1}.zero_g1
|
| 23 |
+
checkpoints/l3_blumodelcp_run1_step_400_rank{0,1}.{ckpt,zero_g1}
|
| 24 |
+
^ l3, the eval baseline. It was the ONE genuine gap found in the final
|
| 25 |
+
audit: HF only had lenbucket_l1_blumodelcp_run1_step_400.ckpt (5.23
|
| 26 |
+
GB), a DIFFERENT file from local blumodelcp_run1_step_400.ckpt (8.18
|
| 27 |
+
GB). l3 appears in every eval table, so it is now backed up.
|
| 28 |
+
logs/misc_local_artifacts_2026-09-09.tgz 5.4 MB: b3_logs, setup_logs,
|
| 29 |
+
mspoe_2026-08-21, mspoe_run100_native, probe_run4,
|
| 30 |
+
context_rot_2026-08-19, CP_BluScreipt_Training_logs, Tests, Scripts,
|
| 31 |
+
context_parallel, graph.ipynb, teardown_logs. These were the local
|
| 32 |
+
dirs with no HF copy. decide_len_{4096,8192,16360} were already on HF
|
| 33 |
+
(26 files each) -- that is the MATCHED decision eval whose l2 numbers
|
| 34 |
+
the 32k results doc cites.
|
| 35 |
|
| 36 |
TO RESUME l3long LATER
|
| 37 |
Pull the four step_2450 files, restore them as
|
|
|
|
| 44 |
2. ext_32k.sh (parent l2, b3 data) -- written, never run. NOTE a DIFFERENT
|
| 45 |
32k extension (ext32k_b4, b4 data) WAS run and evaluated earlier today;
|
| 46 |
see docs/EXT32K_RESULTS_2026-09-09.md and logs/ext32k/.
|
| 47 |
+
3. Source changes: user confirmed pushed to git 2026-09-09. Source tarballs
|
| 48 |
+
on HF remain as a second copy.
|
|
|
|
|
|
|
| 49 |
|
| 50 |
NOT UPLOADED (deliberate, re-derivable)
|
| 51 |
lendata/ 3.3 GB regenerate with fetch_eval_assets.sh
|