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": 76400.34816796797,
-      "required_pages_per_second_upper_estimate": 0.005285553359491259
+      "required_token_positions_per_second": 75831.03642158526,
+      "required_pages_per_second_upper_estimate": 0.0052461670518912555
     },
     {
       "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": 9550.043520995996,
-      "required_pages_per_second_upper_estimate": 0.003998403794654638
+      "required_token_positions_per_second": 9478.879552698158,
+      "required_pages_per_second_upper_estimate": 0.0039686089272009785
     },
     {
       "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": 19100.087041991992,
-      "required_pages_per_second_upper_estimate": 0.003978044427679064
+      "required_token_positions_per_second": 18957.759105396315,
+      "required_pages_per_second_upper_estimate": 0.00394840127192628
     },
     {
       "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": 28650.13056298799,
-      "required_pages_per_second_upper_estimate": 0.0017175315810347483
+      "required_token_positions_per_second": 28436.638658094475,
+      "required_pages_per_second_upper_estimate": 0.001704733067319646
     },
     {
       "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": 28650.13056298799,
-      "required_pages_per_second_upper_estimate": 0.0022274216326660266
+      "required_token_positions_per_second": 28436.638658094475,
+      "required_pages_per_second_upper_estimate": 0.0022108235761122033
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28650.13056298799,
-      "required_pages_per_second_upper_estimate": 0.367493112748528
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28436.638658094475,
+      "required_pages_per_second_upper_estimate": 0.36475466782230187
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1017707091640574,
-    "producer_p95_seconds": 0.2479517299998406,
-    "gpu_update_seconds": 0.9650217802398917,
-    "producer_batches_per_second": 9.826009941504587,
-    "gpu_batches_per_second": 1.0362460417747392,
-    "producer_headroom": 9.482313606405631,
+    "producer_batch_seconds": 0.12024813167960247,
+    "producer_p95_seconds": 0.26581320299919753,
+    "gpu_update_seconds": 0.9722668115744408,
+    "producer_batches_per_second": 8.316137523570594,
+    "gpu_batches_per_second": 1.0285242570201993,
+    "producer_headroom": 8.085504514656549,
     "sustainable": true,
-    "buffer_seconds": 60.796372155113175,
+    "buffer_seconds": 60.28054231761533,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311488.64907820814,
-    "upload_bytes_per_second": 4807090.7530356515,
+    "generation_bytes_per_second": 311974.50051060936,
+    "upload_bytes_per_second": 4805509.442904458,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.83821300754245,
+    "estimated_full_shards_per_hour": 66.94246541489325,
     "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": [
-    {
-      "seconds": 1.110015972999463,
-      "bytes": 6170503,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.4518775689994072,
       "bytes": 12256015,
@@ -562,6 +556,12 @@
       "bytes": 6282079,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.4015076579999004,
+      "bytes": 7431338,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -159,3 +159,4 @@
 {"step": 735900, "loss": 0.5330012440681458, "lr": 5e-05, "grad_norm": 0.2299289590874985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107715457366, "tokens_per_second": 151553.5385937623, "data_wait_s": 1.3471000784193166e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 736000, "loss": 0.5668315291404724, "lr": 5e-05, "grad_norm": 0.27227980005007696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107730231472, "tokens_per_second": 152181.08599206997, "data_wait_s": 1.8500000805943273e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 736100, "loss": 0.5431311130523682, "lr": 5e-05, "grad_norm": 0.24596208616680798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107744761546, "tokens_per_second": 149158.29127311741, "data_wait_s": 1.95359989447752e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 736200, "loss": 0.5547208189964294, "lr": 5e-05, "grad_norm": 0.22673007027056988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107759247246, "tokens_per_second": 146478.99680238252, "data_wait_s": 1.1204001566511579e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e3321b3959fe5a81bff167515bdc78e7729cda30f91d9283cd36babd75b04c01
-size 343512
+oid sha256:59fab8b7a94101519cb642fc43accce8095cd07250311a87d6e3b6fde69a7ec0
+size 345630