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": 77368.00353588616,
-      "required_pages_per_second_upper_estimate": 0.005352498003113631
+      "required_token_positions_per_second": 76898.91960811496,
+      "required_pages_per_second_upper_estimate": 0.00532004568856575
     },
     {
       "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": 9671.00044198577,
-      "required_pages_per_second_upper_estimate": 0.004049045931605321
+      "required_token_positions_per_second": 9612.36495101437,
+      "required_pages_per_second_upper_estimate": 0.004024496475983883
     },
     {
       "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": 19342.00088397154,
-      "required_pages_per_second_upper_estimate": 0.0040284287012663745
+      "required_token_positions_per_second": 19224.72990202874,
+      "required_pages_per_second_upper_estimate": 0.004004004248371454
     },
     {
       "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": 29013.001325957313,
-      "required_pages_per_second_upper_estimate": 0.001739285129203186
+      "required_token_positions_per_second": 28837.09485304311,
+      "required_pages_per_second_upper_estimate": 0.0017287398047456115
     },
     {
       "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": 29013.001325957313,
-      "required_pages_per_second_upper_estimate": 0.002255633238387086
+      "required_token_positions_per_second": 28837.09485304311,
+      "required_pages_per_second_upper_estimate": 0.0022419572838488105
     },
     {
       "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": 29013.001325957313,
-      "required_pages_per_second_upper_estimate": 0.372147629275629
+      "required_token_positions_per_second": 28837.09485304311,
+      "required_pages_per_second_upper_estimate": 0.3698912899147408
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08710982541404633,
-    "producer_p95_seconds": 0.20158138499937195,
-    "gpu_update_seconds": 0.952952081357537,
-    "producer_batches_per_second": 11.479761269717246,
-    "gpu_batches_per_second": 1.0493707076807477,
-    "producer_headroom": 10.93966239546469,
+    "producer_batch_seconds": 0.08160499621858719,
+    "producer_p95_seconds": 0.2274012510006287,
+    "gpu_update_seconds": 0.9587650954750171,
+    "producer_batches_per_second": 12.254151661515913,
+    "gpu_batches_per_second": 1.0430083497194411,
+    "producer_headroom": 11.748852887718643,
     "sustainable": true,
-    "buffer_seconds": 60.03598112552483,
+    "buffer_seconds": 60.40220101492608,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 396319.87848988135,
-    "upload_bytes_per_second": 4277772.467709752,
+    "generation_bytes_per_second": 412124.554688308,
+    "upload_bytes_per_second": 4310662.089848267,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.04102006933527,
+    "estimated_full_shards_per_hour": 88.43233566748552,
     "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": 0.9990272300001379,
-      "bytes": 2457257,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.5623461810000663,
       "bytes": 6316246,
@@ -562,6 +556,12 @@
       "bytes": 13590477,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.1102066319999722,
+      "bytes": 6253510,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -153,3 +153,4 @@
 {"step": 700300, "loss": 0.49784529209136963, "lr": 8e-05, "grad_norm": 0.21931447414439897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102504529645, "tokens_per_second": 146519.1007752696, "data_wait_s": 1.7926000509760343e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 700400, "loss": 0.551393985748291, "lr": 8e-05, "grad_norm": 0.21495483709769558, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102519086586, "tokens_per_second": 152066.0889579606, "data_wait_s": 1.5417999748024158e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 700500, "loss": 0.5333302617073059, "lr": 8e-05, "grad_norm": 0.19961868484414885, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102533770750, "tokens_per_second": 152591.01016817932, "data_wait_s": 1.4356000974657945e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 700600, "loss": 0.5083737969398499, "lr": 8e-05, "grad_norm": 0.20765770297318153, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102548441783, "tokens_per_second": 153083.31381029068, "data_wait_s": 1.3606999345938675e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4dd76c42a778f8b2409e1ad7af9f85989bf498a6d067822da4a2a476baa1de97
-size 330804
+oid sha256:a1bafb8dda698f3e209290d7fcc1abb85412d12de3f8c88de2baab8342d73002
+size 332922