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": 77062.74423880134,
-      "required_pages_per_second_upper_estimate": 0.005331379456642149
+      "required_token_positions_per_second": 75996.9672047794,
+      "required_pages_per_second_upper_estimate": 0.005257646528485087
     },
     {
       "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": 9632.843029850168,
-      "required_pages_per_second_upper_estimate": 0.004033070219961518
+      "required_token_positions_per_second": 9499.620900597425,
+      "required_pages_per_second_upper_estimate": 0.0039772928964378045
     },
     {
       "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": 19265.686059700336,
-      "required_pages_per_second_upper_estimate": 0.004012534335927937
+      "required_token_positions_per_second": 18999.24180119485,
+      "required_pages_per_second_upper_estimate": 0.003957041023488861
     },
     {
       "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": 28898.529089550502,
-      "required_pages_per_second_upper_estimate": 0.0017324226934195821
+      "required_token_positions_per_second": 28498.862701792274,
+      "required_pages_per_second_upper_estimate": 0.0017084633037287155
     },
     {
       "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": 28898.529089550502,
-      "required_pages_per_second_upper_estimate": 0.002246733525516612
+      "required_token_positions_per_second": 28498.862701792274,
+      "required_pages_per_second_upper_estimate": 0.002215661221815064
     },
     {
       "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": 28898.529089550502,
-      "required_pages_per_second_upper_estimate": 0.37067930233771373
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28498.862701792274,
+      "required_pages_per_second_upper_estimate": 0.36555281104388443
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.22608394276570465,
-    "producer_p95_seconds": 0.18562298300003022,
-    "gpu_update_seconds": 0.9567268947953934,
-    "producer_batches_per_second": 4.423135883808963,
-    "gpu_batches_per_second": 1.045230363482006,
-    "producer_headroom": 4.231733059374627,
+    "producer_batch_seconds": 0.07469833057813702,
+    "producer_p95_seconds": 0.21088980000058655,
+    "gpu_update_seconds": 0.9701439769475866,
+    "producer_batches_per_second": 13.387180038166525,
+    "gpu_batches_per_second": 1.030774837304408,
+    "producer_headroom": 12.987492082340216,
     "sustainable": true,
-    "buffer_seconds": 60.27379437210978,
+    "buffer_seconds": 60.14892657075037,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354646.07434368425,
-    "upload_bytes_per_second": 4818867.077246846,
+    "generation_bytes_per_second": 351582.3807806344,
+    "upload_bytes_per_second": 4833566.687807606,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.09879181607147,
+    "estimated_full_shards_per_hour": 75.44139449657702,
     "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": [
@@ -532,6 +532,12 @@
       "bytes": 7724848,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 0.9524561609996454,
+      "bytes": 5747185,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -70,3 +70,4 @@
 {"step": 404100, "loss": 0.533147394657135, "lr": 0.0003, "grad_norm": 0.15824127040319796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59149658570, "tokens_per_second": 152590.37667120722, "data_wait_s": 1.0043000656878576e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 404200, "loss": 0.5693563222885132, "lr": 0.0003, "grad_norm": 0.15882381498968365, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59164350744, "tokens_per_second": 152735.23199207423, "data_wait_s": 1.4741000086360145e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 404300, "loss": 0.5971654653549194, "lr": 0.0003, "grad_norm": 0.15261377341042853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59179156258, "tokens_per_second": 154304.80222531557, "data_wait_s": 1.5270000403688755e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 404400, "loss": 0.5785234570503235, "lr": 0.0003, "grad_norm": 0.15924016359875773, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59193720644, "tokens_per_second": 150942.61431131698, "data_wait_s": 8.276999324152712e-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 sha256:94151b8d5ae1ad099b4b5c6d62b02fbdf9a132209ba6231de102059d5c99e45d
-size 153444
+oid sha256:dd3402bb8d1bfdd02fad85669dfd5ad41d6cb96f5d1c0f1ed068de1f7e8fec22
+size 155562