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": 75832.0306972349,
-      "required_pages_per_second_upper_estimate": 0.005246235838187735
+      "required_token_positions_per_second": 77226.85627128536,
+      "required_pages_per_second_upper_estimate": 0.005342733107841772
     },
     {
       "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": 9479.003837154363,
-      "required_pages_per_second_upper_estimate": 0.003968660962507437
+      "required_token_positions_per_second": 9653.35703391067,
+      "required_pages_per_second_upper_estimate": 0.004041659005080533
     },
     {
       "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": 18958.007674308727,
-      "required_pages_per_second_upper_estimate": 0.003948453042275532
+      "required_token_positions_per_second": 19306.71406782134,
+      "required_pages_per_second_upper_estimate": 0.004021079388038208
     },
     {
       "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": 28437.01151146309,
-      "required_pages_per_second_upper_estimate": 0.0017047554193097803
+      "required_token_positions_per_second": 28960.07110173201,
+      "required_pages_per_second_upper_estimate": 0.0017361120430806518
     },
     {
       "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": 28437.01151146309,
-      "required_pages_per_second_upper_estimate": 0.0022108525638216046
+      "required_token_positions_per_second": 28960.07110173201,
+      "required_pages_per_second_upper_estimate": 0.0022515181462690204
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28437.01151146309,
-      "required_pages_per_second_upper_estimate": 0.3647594503849757
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28960.07110173201,
+      "required_pages_per_second_upper_estimate": 0.37146869719131376
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.8210969655311828,
-    "producer_p95_seconds": 0.36235123899678,
-    "gpu_update_seconds": 0.9722540636471229,
-    "producer_batches_per_second": 1.2178829565556626,
-    "gpu_batches_per_second": 1.0285377427467841,
-    "producer_headroom": 1.1840916535578154,
+    "producer_batch_seconds": 0.11180079299242607,
+    "producer_p95_seconds": 0.33946742200350855,
+    "gpu_update_seconds": 0.9546937886608455,
+    "producer_batches_per_second": 8.94448038546332,
+    "gpu_batches_per_second": 1.047456275380932,
+    "producer_headroom": 8.539239866800598,
     "sustainable": true,
-    "buffer_seconds": 60.279751946121614,
+    "buffer_seconds": 60.14570868563327,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 279291.4860400919,
+    "generation_bytes_per_second": 279617.68357396417,
     "upload_bytes_per_second": 5136263.988948157,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.92945133115833,
+    "estimated_full_shards_per_hour": 59.999445728437365,
     "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": [

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -419,3 +419,4 @@
 {"step": 439000, "loss": 0.6220097541809082, "lr": 0.0003, "grad_norm": 0.18819345086694877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64256836105, "tokens_per_second": 151936.96073377918, "data_wait_s": 1.0144998668693006e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 439100, "loss": 0.5693238973617554, "lr": 0.0003, "grad_norm": 0.1412137916417348, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64271512797, "tokens_per_second": 152294.6378369886, "data_wait_s": 8.383998647332191e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 439200, "loss": 0.6370528936386108, "lr": 0.0003, "grad_norm": 0.1636667113696795, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64286135171, "tokens_per_second": 120852.7214850648, "data_wait_s": 1.022200012812391e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 439300, "loss": 0.6133816242218018, "lr": 0.0003, "grad_norm": 0.1552689703390891, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64300835505, "tokens_per_second": 152617.73701016037, "data_wait_s": 9.742005204316229e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0b4257f17fbfc49ba00db69100e09f14b3756b39a14ad068c33847d5c6957c1e
-size 898107
+oid sha256:484b5bea6c6d4c95f92277bf87f2ea953a89b25bb5d656d2af8ffde631b82e87
+size 900225