Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77431.425011619,
-      "required_pages_per_second_upper_estimate": 0.005356885647962926
+      "required_token_positions_per_second": 77345.5715405323,
+      "required_pages_per_second_upper_estimate": 0.00535094610562565
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9678.928126452374,
-      "required_pages_per_second_upper_estimate": 0.004052365087542611
+      "required_token_positions_per_second": 9668.196442566537,
+      "required_pages_per_second_upper_estimate": 0.004047871955602654
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19357.85625290475,
-      "required_pages_per_second_upper_estimate": 0.004031730956480965
+      "required_token_positions_per_second": 19336.392885133075,
+      "required_pages_per_second_upper_estimate": 0.004027260703001197
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29036.784379357123,
-      "required_pages_per_second_upper_estimate": 0.0017407108869398847
+      "required_token_positions_per_second": 29004.58932769961,
+      "required_pages_per_second_upper_estimate": 0.0017387808427520143
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29036.784379357123,
-      "required_pages_per_second_upper_estimate": 0.002257482266178325
+      "required_token_positions_per_second": 29004.58932769961,
+      "required_pages_per_second_upper_estimate": 0.0022549792425230197
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29036.784379357123,
-      "required_pages_per_second_upper_estimate": 0.3724526927483886
+      "required_token_positions_per_second": 29004.58932769961,
+      "required_pages_per_second_upper_estimate": 0.37203972919407924
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08042578320311833,
-    "producer_p95_seconds": 0.1762838639997426,
-    "gpu_update_seconds": 0.9521715503613258,
-    "producer_batches_per_second": 12.433823584589318,
-    "gpu_batches_per_second": 1.0502309164987385,
-    "producer_headroom": 11.839133079457628,
+    "producer_batch_seconds": 0.07859821138279699,
+    "producer_p95_seconds": 0.17063264199987316,
+    "gpu_update_seconds": 0.9532284593871475,
+    "producer_batches_per_second": 12.722935832848645,
+    "gpu_batches_per_second": 1.0490664542715427,
+    "producer_headroom": 12.127864522827847,
     "sustainable": true,
-    "buffer_seconds": 60.93897922312485,
+    "buffer_seconds": 60.05339294139029,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 322099.35823175503,
-    "upload_bytes_per_second": 3526207.4037426272,
+    "generation_bytes_per_second": 321694.6399920697,
+    "upload_bytes_per_second": 3612232.7737536477,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.11502418722618,
+    "estimated_full_shards_per_hour": 69.02818107434814,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
@@ -280,6 +280,12 @@
       "bytes": 9679861,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.4209248310007752,
+      "bytes": 7838674,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -60,3 +60,4 @@
 {"step": 931000, "loss": 0.5477752685546875, "lr": 3e-05, "grad_norm": 0.33733932884986956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136272617374, "tokens_per_second": 155744.74330611492, "data_wait_s": 8.26599989522947e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 931100, "loss": 0.5290407538414001, "lr": 3e-05, "grad_norm": 0.33868731431852284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136287143891, "tokens_per_second": 151374.79390160504, "data_wait_s": 1.2568000784085598e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 931200, "loss": 0.48341676592826843, "lr": 3e-05, "grad_norm": 0.3590094845794258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136301676493, "tokens_per_second": 152018.1240500222, "data_wait_s": 1.1115000233985484e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 931300, "loss": 0.5140773057937622, "lr": 3e-05, "grad_norm": 0.3326559928877842, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136316230553, "tokens_per_second": 152114.58814402384, "data_wait_s": 1.170999985333765e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:776cbb84f0ec432e0242e385c6eba9365d81aeed4beafa63827ca4ab14148e7a
-size 132264
+oid sha256:6c30f09a3148e4686f37fcb6233941501ec5c07c799e2dfd5748ed7ccdbe16b8
+size 134382