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": 75542.71858122038,
-      "required_pages_per_second_upper_estimate": 0.005226220554705254
+      "required_token_positions_per_second": 76173.66152088203,
+      "required_pages_per_second_upper_estimate": 0.005269870651260365
     },
     {
       "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": 9442.839822652548,
-      "required_pages_per_second_upper_estimate": 0.003953519844826025
+      "required_token_positions_per_second": 9521.707690110254,
+      "required_pages_per_second_upper_estimate": 0.003986540173982213
     },
     {
       "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": 18885.679645305096,
-      "required_pages_per_second_upper_estimate": 0.003933389021252469
+      "required_token_positions_per_second": 19043.415380220507,
+      "required_pages_per_second_upper_estimate": 0.003966241215064287
     },
     {
       "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": 28328.519467957645,
-      "required_pages_per_second_upper_estimate": 0.0016982514869593873
+      "required_token_positions_per_second": 28565.12307033076,
+      "required_pages_per_second_upper_estimate": 0.001712435511648873
     },
     {
       "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": 28328.519467957645,
-      "required_pages_per_second_upper_estimate": 0.0022024177846451227
+      "required_token_positions_per_second": 28565.12307033076,
+      "required_pages_per_second_upper_estimate": 0.002220812674020371
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28328.519467957645,
-      "required_pages_per_second_upper_estimate": 0.3633678309405683
+      "required_token_positions_per_second": 28565.12307033076,
+      "required_pages_per_second_upper_estimate": 0.36640272790665535
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.3082125501796895,
-    "producer_p95_seconds": 0.24971950400140486,
-    "gpu_update_seconds": 0.9759775844012116,
-    "producer_batches_per_second": 3.244514214028581,
-    "gpu_batches_per_second": 1.0246136960343477,
-    "producer_headroom": 3.1665731451630106,
+    "producer_batch_seconds": 0.09291623697657769,
+    "producer_p95_seconds": 0.24399043499943218,
+    "gpu_update_seconds": 0.9678936068970299,
+    "producer_batches_per_second": 10.762381608847114,
+    "gpu_batches_per_second": 1.0331714073470328,
+    "producer_headroom": 10.416840354189294,
     "sustainable": true,
-    "buffer_seconds": 60.51061023287512,
+    "buffer_seconds": 60.00940362761585,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 294363.02759634133,
-    "upload_bytes_per_second": 4772045.511904262,
+    "generation_bytes_per_second": 295955.50260809954,
+    "upload_bytes_per_second": 4723481.286954697,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.16345330159836,
+    "estimated_full_shards_per_hour": 63.50516136820068,
     "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": 1.1991475859995262,
-      "bytes": 11620692,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 0.9693775400010054,
       "bytes": 6039517,
@@ -562,6 +556,12 @@
       "bytes": 5945434,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1649709810008062,
+      "bytes": 7159686,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -188,3 +188,4 @@
 {"step": 415900, "loss": 0.5758548378944397, "lr": 0.0003, "grad_norm": 0.15755216020757526, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60876652439, "tokens_per_second": 117081.74765913683, "data_wait_s": 9.520001185592264e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 416000, "loss": 0.6045684814453125, "lr": 0.0003, "grad_norm": 0.15525646490416434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60891258285, "tokens_per_second": 151928.47331500877, "data_wait_s": 1.3898999895900488e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 416100, "loss": 0.5677547454833984, "lr": 0.0003, "grad_norm": 0.14811392401979356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60905871071, "tokens_per_second": 151102.96985638948, "data_wait_s": 1.9032999261980876e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 416200, "loss": 0.5876190066337585, "lr": 0.0003, "grad_norm": 0.1630453605535547, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60920512418, "tokens_per_second": 152446.98709928765, "data_wait_s": 1.0996998753398657e-05, "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:c4e345ea71dd58dd66f2ef82c35b4987c9186ae203ec49f2e0d562b61c85444e
-size 405717
+oid sha256:716e33f759e11ceb2abc6a4edb676f9d89a20b1cd5b3d651e82cb6a94e62798a
+size 407835