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": 76362.34617830475,
-      "required_pages_per_second_upper_estimate": 0.0052829242936695935
+      "required_token_positions_per_second": 77066.55555951691,
+      "required_pages_per_second_upper_estimate": 0.005331643132652735
     },
     {
       "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": 9545.293272288094,
-      "required_pages_per_second_upper_estimate": 0.003996414964717113
+      "required_token_positions_per_second": 9633.319444939614,
+      "required_pages_per_second_upper_estimate": 0.004033269685010791
     },
     {
       "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": 19090.58654457619,
-      "required_pages_per_second_upper_estimate": 0.003976065724612317
+      "required_token_positions_per_second": 19266.63888987923,
+      "required_pages_per_second_upper_estimate": 0.004012732785326379
     },
     {
       "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": 28635.879816864282,
-      "required_pages_per_second_upper_estimate": 0.0017166772705642615
+      "required_token_positions_per_second": 28899.95833481884,
+      "required_pages_per_second_upper_estimate": 0.0017325083744910877
     },
     {
       "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": 28635.879816864282,
-      "required_pages_per_second_upper_estimate": 0.002226313699837316
+      "required_token_positions_per_second": 28899.95833481884,
+      "required_pages_per_second_upper_estimate": 0.0022468446430496396
     },
     {
       "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": 28635.879816864282,
-      "required_pages_per_second_upper_estimate": 0.3673103194785121
+      "required_token_positions_per_second": 28899.95833481884,
+      "required_pages_per_second_upper_estimate": 0.37069763516141196
     }
   ],
   "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": 0.35779279762495264,
-    "producer_p95_seconds": 0.20579205499961972,
-    "gpu_update_seconds": 0.9655020267167591,
-    "producer_batches_per_second": 2.7949137227972516,
-    "gpu_batches_per_second": 1.0357306068020935,
-    "producer_headroom": 2.6984948638592288,
+    "producer_batch_seconds": 0.07971982097653552,
+    "producer_p95_seconds": 0.19052149900016957,
+    "gpu_update_seconds": 0.9566795799386855,
+    "producer_batches_per_second": 12.543931832139172,
+    "gpu_batches_per_second": 1.045282057827649,
+    "producer_headroom": 12.00052343595041,
     "sustainable": true,
-    "buffer_seconds": 60.82662768315583,
+    "buffer_seconds": 60.27081353613718,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 333480.12593408546,
-    "upload_bytes_per_second": 4461890.28772334,
+    "generation_bytes_per_second": 296294.29968070536,
+    "upload_bytes_per_second": 4493242.885644496,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.5570720054333,
+    "estimated_full_shards_per_hour": 63.577859333189686,
     "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": [
@@ -544,6 +544,12 @@
       "bytes": 9834674,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.1737758689996554,
+      "bytes": 8190004,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -77,3 +77,4 @@
 {"step": 447900, "loss": 0.5529193878173828, "lr": 0.0003, "grad_norm": 0.15696307109837762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65558740424, "tokens_per_second": 154257.8371211841, "data_wait_s": 1.1615000403253362e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 448000, "loss": 0.6013756990432739, "lr": 0.0003, "grad_norm": 0.16911388717877565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65573386473, "tokens_per_second": 152753.0935244055, "data_wait_s": 1.1083999197580852e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 448100, "loss": 0.5607942938804626, "lr": 0.0003, "grad_norm": 0.17170184886871018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65588091633, "tokens_per_second": 151717.5884138647, "data_wait_s": 1.4432000170927495e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 448200, "loss": 0.6464706659317017, "lr": 0.0003, "grad_norm": 0.15260309219318896, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65602686693, "tokens_per_second": 152285.74856316112, "data_wait_s": 1.0685000233934261e-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:d9369ce52103655f0b9a2b8a619ab74e183dd8441ed110a481dd5bdcbf4de770
-size 168270
+oid sha256:0d7ce90efb6920c7c6339294dd0326e153d4de871471d5472c0fe5de6b2d9dd9
+size 170388