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": 77399.60978839574,
-      "required_pages_per_second_upper_estimate": 0.005354684596998853
+      "required_token_positions_per_second": 77073.84154725683,
+      "required_pages_per_second_upper_estimate": 0.005332147194190403
     },
     {
       "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": 9674.951223549468,
-      "required_pages_per_second_upper_estimate": 0.00405070004134434
+      "required_token_positions_per_second": 9634.230193407104,
+      "required_pages_per_second_upper_estimate": 0.004033650996375538
     },
     {
       "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": 19349.902447098935,
-      "required_pages_per_second_upper_estimate": 0.004030074388487579
+      "required_token_positions_per_second": 19268.460386814208,
+      "required_pages_per_second_upper_estimate": 0.004013112155101831
     },
     {
       "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": 29024.8536706484,
-      "required_pages_per_second_upper_estimate": 0.0017399956591699344
+      "required_token_positions_per_second": 28902.690580221315,
+      "required_pages_per_second_upper_estimate": 0.001732672168431069
     },
     {
       "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": 29024.8536706484,
-      "required_pages_per_second_upper_estimate": 0.0022565547060538634
+      "required_token_positions_per_second": 28902.690580221315,
+      "required_pages_per_second_upper_estimate": 0.0022470570631118046
     },
     {
       "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": 29024.8536706484,
-      "required_pages_per_second_upper_estimate": 0.3722996584272701
+      "required_token_positions_per_second": 28902.690580221315,
+      "required_pages_per_second_upper_estimate": 0.3707326814717784
     }
   ],
   "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": 0.15978182482811576,
-    "producer_p95_seconds": 0.25138754199724644,
-    "gpu_update_seconds": 0.9525629418748541,
-    "producer_batches_per_second": 6.258534104712744,
-    "gpu_batches_per_second": 1.0497993949163917,
-    "producer_headroom": 5.961647658609278,
+    "producer_batch_seconds": 0.07043148302349778,
+    "producer_p95_seconds": 0.2526048450017697,
+    "gpu_update_seconds": 0.9565891425665428,
+    "producer_batches_per_second": 14.198195992357196,
+    "gpu_batches_per_second": 1.0453808803610138,
+    "producer_headroom": 13.581840130320694,
     "sustainable": true,
-    "buffer_seconds": 60.01146533811581,
+    "buffer_seconds": 60.2651159816922,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346973.7765264379,
+    "generation_bytes_per_second": 304690.297907737,
     "upload_bytes_per_second": 5546989.657525898,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.45249530644276,
+    "estimated_full_shards_per_hour": 65.379445103875,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -260,3 +260,4 @@
 {"step": 583100, "loss": 0.5865918397903442, "lr": 0.0003, "grad_norm": 0.1487584477832386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85348969633, "tokens_per_second": 152443.44046201778, "data_wait_s": 1.1420001101214439e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 583200, "loss": 0.579066276550293, "lr": 0.0003, "grad_norm": 0.15166410431059976, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85363615908, "tokens_per_second": 153229.74095765813, "data_wait_s": 1.51149979501497e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 583300, "loss": 0.5727190375328064, "lr": 0.0003, "grad_norm": 0.17853147238948933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85378311432, "tokens_per_second": 152224.17760109832, "data_wait_s": 9.898998541757464e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 583400, "loss": 0.6119234561920166, "lr": 0.0003, "grad_norm": 0.1917100880345442, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85392954574, "tokens_per_second": 152515.2701479299, "data_wait_s": 1.4774999726796523e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:93459dd4639e9ee42a60c110e5fc735fa998c9e9cb4211d3aa38e67764d99c74
-size 558996
+oid sha256:bb079796ab034eb8d2104a47bd41573b8f027e9c9ed8c4a3d028f4c08a338274
+size 561114