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": 76313.90078282343,
-      "required_pages_per_second_upper_estimate": 0.0052795727288013905
+      "required_token_positions_per_second": 76210.58668655438,
+      "required_pages_per_second_upper_estimate": 0.0052724252198997675
     },
     {
       "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": 9539.237597852929,
-      "required_pages_per_second_upper_estimate": 0.003993879580287971
+      "required_token_positions_per_second": 9526.323335819297,
+      "required_pages_per_second_upper_estimate": 0.003988472648454948
     },
     {
       "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": 19078.475195705858,
-      "required_pages_per_second_upper_estimate": 0.003973543250040375
+      "required_token_positions_per_second": 19052.646671638595,
+      "required_pages_per_second_upper_estimate": 0.003968163849621149
     },
     {
       "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": 28617.71279355879,
-      "required_pages_per_second_upper_estimate": 0.0017155881852565357
+      "required_token_positions_per_second": 28578.97000745789,
+      "required_pages_per_second_upper_estimate": 0.001713265614386596
     },
     {
       "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": 28617.71279355879,
-      "required_pages_per_second_upper_estimate": 0.0022249012936836044
+      "required_token_positions_per_second": 28578.97000745789,
+      "required_pages_per_second_upper_estimate": 0.002221889212475766
     },
     {
       "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": 28617.71279355879,
-      "required_pages_per_second_upper_estimate": 0.3670772924097791
+      "required_token_positions_per_second": 28578.97000745789,
+      "required_pages_per_second_upper_estimate": 0.3665803415484395
     }
   ],
   "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": 0.0875303213515295,
-    "producer_p95_seconds": 0.15692202100035502,
-    "gpu_update_seconds": 0.9661149442461017,
-    "producer_batches_per_second": 11.424612460679903,
-    "gpu_batches_per_second": 1.0350735240725835,
-    "producer_headroom": 11.037488830483083,
+    "producer_batch_seconds": 0.2699154584531058,
+    "producer_p95_seconds": 0.1620461559996329,
+    "gpu_update_seconds": 0.9674246480116342,
+    "producer_batches_per_second": 3.704863758937825,
+    "gpu_batches_per_second": 1.033672236959559,
+    "producer_headroom": 3.5841765179214855,
     "sustainable": true,
-    "buffer_seconds": 60.865241487504406,
+    "buffer_seconds": 60.94775282473295,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 518533.10500759823,
+    "generation_bytes_per_second": 495563.8410038934,
     "upload_bytes_per_second": 4332706.564006769,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.2651335017296,
+    "estimated_full_shards_per_hour": 106.33646414363481,
     "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

@@ -39,3 +39,4 @@
 {"step": 444100, "loss": 0.5636587142944336, "lr": 0.0003, "grad_norm": 0.1544770824898031, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65002947538, "tokens_per_second": 150526.41377154645, "data_wait_s": 1.225300002261065e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 444200, "loss": 0.5759021639823914, "lr": 0.0003, "grad_norm": 0.17241649940589882, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65017382051, "tokens_per_second": 150116.48436633035, "data_wait_s": 1.4088000170886517e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 444300, "loss": 0.5492525696754456, "lr": 0.0003, "grad_norm": 0.20942501199811075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65032080794, "tokens_per_second": 153140.43664036572, "data_wait_s": 1.4805000319029205e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 444400, "loss": 0.5989384055137634, "lr": 0.0003, "grad_norm": 0.16974348781045961, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65046721194, "tokens_per_second": 152128.5663943477, "data_wait_s": 1.2406000678311102e-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 sha256:5f83a2a5b4ada3b0c9d041ca5d83840b0b15ce3c35cc6f019eac9743ce8de20a
-size 87003
+oid sha256:9e55aba44a3a9ca33ad6110c021e1733089f6e986ac9f3efb765babcf922b9fb
+size 89121