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": 76982.93247032835,
-      "required_pages_per_second_upper_estimate": 0.005325857893310371
+      "required_token_positions_per_second": 76406.2405888879,
+      "required_pages_per_second_upper_estimate": 0.0052859610108428
     },
     {
       "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": 9622.866558791044,
-      "required_pages_per_second_upper_estimate": 0.004028893279861469
+      "required_token_positions_per_second": 9550.780073610988,
+      "required_pages_per_second_upper_estimate": 0.003998712173853568
     },
     {
       "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": 19245.733117582087,
-      "required_pages_per_second_upper_estimate": 0.004008378664279175
+      "required_token_positions_per_second": 19101.560147221975,
+      "required_pages_per_second_upper_estimate": 0.003978351236650072
     },
     {
       "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": 28868.599676373135,
-      "required_pages_per_second_upper_estimate": 0.0017306284707991667
+      "required_token_positions_per_second": 28652.340220832964,
+      "required_pages_per_second_upper_estimate": 0.0017176640466486026
     },
     {
       "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": 28868.599676373135,
-      "required_pages_per_second_upper_estimate": 0.0022444066452876475
+      "required_token_positions_per_second": 28652.340220832964,
+      "required_pages_per_second_upper_estimate": 0.0022275934238442157
     },
     {
       "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": 28868.599676373135,
-      "required_pages_per_second_upper_estimate": 0.3702953999611746
+      "required_token_positions_per_second": 28652.340220832964,
+      "required_pages_per_second_upper_estimate": 0.36752145586681756
     }
   ],
   "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": 0.06192021532029557,
-    "producer_p95_seconds": 0.14521822000006068,
-    "gpu_update_seconds": 0.9577187778397126,
-    "producer_batches_per_second": 16.149814641749643,
-    "gpu_batches_per_second": 1.0441478470910421,
-    "producer_headroom": 15.466980741034366,
+    "producer_batch_seconds": 0.16272236082813407,
+    "producer_p95_seconds": 0.14889653499994893,
+    "gpu_update_seconds": 0.9649473581183182,
+    "producer_batches_per_second": 6.1454368957699135,
+    "gpu_batches_per_second": 1.0363259628484145,
+    "producer_headroom": 5.930023097056017,
     "sustainable": true,
-    "buffer_seconds": 60.33628300390189,
+    "buffer_seconds": 60.79168356145405,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 414313.41611709446,
-    "upload_bytes_per_second": 5253481.751966967,
+    "generation_bytes_per_second": 422396.62886957865,
+    "upload_bytes_per_second": 5499815.520343313,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.90201437601687,
+    "estimated_full_shards_per_hour": 90.63648366513748,
     "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": [
@@ -322,6 +322,12 @@
       "bytes": 1131427,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.1826321499997903,
+      "bytes": 12996691,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -26,3 +26,4 @@
 {"step": 559700, "loss": 0.5579128265380859, "lr": 0.0003, "grad_norm": 0.15052356989310225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81922378235, "tokens_per_second": 152137.26047137222, "data_wait_s": 1.77590000021155e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 559800, "loss": 0.6557629704475403, "lr": 0.0003, "grad_norm": 0.20263330391709677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81936917792, "tokens_per_second": 151485.46904864797, "data_wait_s": 1.386499980071676e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 559900, "loss": 0.5928296446800232, "lr": 0.0003, "grad_norm": 0.17015354657335416, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81951488575, "tokens_per_second": 151433.72141466296, "data_wait_s": 1.1948000064876396e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 560000, "loss": 0.5581508874893188, "lr": 0.0003, "grad_norm": 0.15723244372163023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81966036542, "tokens_per_second": 151563.19488607012, "data_wait_s": 8.724000053916825e-06, "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:edcb36b878a31173e837323384110508e49ee76c3d60bc32a392d1af35c9c301
-size 59469
+oid sha256:89524613046ccef54dc19f05967a82584cb8ec82f6626bf99954e67b617cf458
+size 61587