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": 76515.6069716057,
-      "required_pages_per_second_upper_estimate": 0.005293527230964187
+      "required_token_positions_per_second": 77118.47060954582,
+      "required_pages_per_second_upper_estimate": 0.005335234736273247
     },
     {
       "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": 9564.450871450712,
-      "required_pages_per_second_upper_estimate": 0.004004435851430338
+      "required_token_positions_per_second": 9639.808826193228,
+      "required_pages_per_second_upper_estimate": 0.004035986653427989
     },
     {
       "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": 19128.901742901424,
-      "required_pages_per_second_upper_estimate": 0.003984045769983735
+      "required_token_positions_per_second": 19279.617652386456,
+      "required_pages_per_second_upper_estimate": 0.00401543591928365
     },
     {
       "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": 28693.352614352134,
-      "required_pages_per_second_upper_estimate": 0.0017201226770177777
+      "required_token_positions_per_second": 28919.42647857968,
+      "required_pages_per_second_upper_estimate": 0.0017336754599834158
     },
     {
       "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": 28693.352614352134,
-      "required_pages_per_second_upper_estimate": 0.0022307819570458764
+      "required_token_positions_per_second": 28919.42647857968,
+      "required_pages_per_second_upper_estimate": 0.0022483582055957303
     },
     {
       "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": 28693.352614352134,
-      "required_pages_per_second_upper_estimate": 0.3680475188152042
+      "required_token_positions_per_second": 28919.42647857968,
+      "required_pages_per_second_upper_estimate": 0.37094735160631187
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1265542446328709,
-    "producer_p95_seconds": 0.30294875600156956,
-    "gpu_update_seconds": 0.9635681257466839,
-    "producer_batches_per_second": 7.901749980026054,
-    "gpu_batches_per_second": 1.0378093393501207,
-    "producer_headroom": 7.6138744183726015,
+    "producer_batch_seconds": 0.5661119113437678,
+    "producer_p95_seconds": 0.31048455400014063,
+    "gpu_update_seconds": 0.9560355569457294,
+    "producer_batches_per_second": 1.7664351870398227,
+    "gpu_batches_per_second": 1.0459862007588137,
+    "producer_headroom": 1.6887748478501505,
     "sustainable": true,
-    "buffer_seconds": 60.70479192204109,
+    "buffer_seconds": 60.23024008758095,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 447425.2385677717,
+    "generation_bytes_per_second": 375913.54090569075,
     "upload_bytes_per_second": 5364158.44416546,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.00704067015518,
+    "estimated_full_shards_per_hour": 80.66229505899469,
     "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-v1/1790868774355361490-de221e1d/train.jsonl

@@ -279,3 +279,4 @@
 {"step": 468100, "loss": 0.5713925361633301, "lr": 0.0003, "grad_norm": 0.13694000479246665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68515960376, "tokens_per_second": 149767.124380907, "data_wait_s": 1.3064000086160377e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 468200, "loss": 0.5405552983283997, "lr": 0.0003, "grad_norm": 0.168116357562067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68530560082, "tokens_per_second": 151650.48431861398, "data_wait_s": 1.3097000191919506e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 468300, "loss": 0.5770984888076782, "lr": 0.0003, "grad_norm": 0.16690113001715673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68545227487, "tokens_per_second": 152071.72160576063, "data_wait_s": 1.233999864780344e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 468400, "loss": 0.5596917271614075, "lr": 0.0003, "grad_norm": 0.13621850775012642, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68559753609, "tokens_per_second": 150883.02814768994, "data_wait_s": 1.5162997442530468e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:20df66c31213fc9ca93626f76fa165cbd3303dde969a5314bed1cd55fc076c0a
-size 599238
+oid sha256:8d68f70e2578424723ecb69373ea1ced2399aedacb05289dc83f262659e6649a
+size 601356