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": 77026.21189586056,
-      "required_pages_per_second_upper_estimate": 0.0053288520643907916
+      "required_token_positions_per_second": 76616.74472394725,
+      "required_pages_per_second_upper_estimate": 0.0053005241753430925
     },
     {
       "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": 9628.27648698257,
-      "required_pages_per_second_upper_estimate": 0.004031158303823114
+      "required_token_positions_per_second": 9577.093090493407,
+      "required_pages_per_second_upper_estimate": 0.00400972888454396
     },
     {
       "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": 19256.55297396514,
-      "required_pages_per_second_upper_estimate": 0.0040106321550249634
+      "required_token_positions_per_second": 19154.186180986813,
+      "required_pages_per_second_upper_estimate": 0.003989311851641449
     },
     {
       "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": 28884.82946094771,
-      "required_pages_per_second_upper_estimate": 0.0017316014216029664
+      "required_token_positions_per_second": 28731.27927148022,
+      "required_pages_per_second_upper_estimate": 0.001722396322201954
     },
     {
       "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": 28884.82946094771,
-      "required_pages_per_second_upper_estimate": 0.0022456684396510427
+      "required_token_positions_per_second": 28731.27927148022,
+      "required_pages_per_second_upper_estimate": 0.002233730587815851
     },
     {
       "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": 28884.82946094771,
-      "required_pages_per_second_upper_estimate": 0.3705035782115123
+      "required_token_positions_per_second": 28731.27927148022,
+      "required_pages_per_second_upper_estimate": 0.3685340012503716
     }
   ],
   "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": 0.09560513774994917,
-    "producer_p95_seconds": 0.17555343200001516,
-    "gpu_update_seconds": 0.9571806555887787,
-    "producer_batches_per_second": 10.459688919809455,
-    "gpu_batches_per_second": 1.0447348618687684,
-    "producer_headroom": 10.0118118975179,
+    "producer_batch_seconds": 0.630039377312464,
+    "producer_p95_seconds": 0.1873292529999162,
+    "gpu_update_seconds": 0.9622961699253146,
+    "producer_batches_per_second": 1.5872023813268048,
+    "gpu_batches_per_second": 1.0391811079094408,
+    "producer_headroom": 1.527358772447123,
     "sustainable": true,
-    "buffer_seconds": 60.30238130209306,
+    "buffer_seconds": 60.62465870529482,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 396458.53455994703,
-    "upload_bytes_per_second": 3748060.2117251144,
+    "generation_bytes_per_second": 343155.0291015873,
+    "upload_bytes_per_second": 3728041.206919366,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.07077243422326,
+    "estimated_full_shards_per_hour": 73.6330810049602,
     "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": [
@@ -370,6 +370,12 @@
       "bytes": 6092133,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.8937167260000933,
+      "bytes": 2214439,
+      "pages": 24,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -55,3 +55,4 @@
 {"step": 530500, "loss": 0.556228756904602, "lr": 0.0003, "grad_norm": 0.13982951517325398, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77647792046, "tokens_per_second": 150454.706173649, "data_wait_s": 1.3650999790115748e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 530600, "loss": 0.5936905741691589, "lr": 0.0003, "grad_norm": 0.1371403233082419, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77662288024, "tokens_per_second": 151018.46994342774, "data_wait_s": 1.4273000488174148e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 530700, "loss": 0.5852376222610474, "lr": 0.0003, "grad_norm": 0.15602347934073346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77676844345, "tokens_per_second": 130556.67780658962, "data_wait_s": 1.9743999473575968e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 530800, "loss": 0.5456879734992981, "lr": 0.0003, "grad_norm": 0.15971735096781994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77691532372, "tokens_per_second": 153204.04569964865, "data_wait_s": 1.313399934588233e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0a0a9975aca1206aa09009c143855b890d68ce91f142e86548bb9f91d9ad2566
-size 121674
+oid sha256:4c8ff7d158c94978a6781e0ed5c02ee230660092e4eae39702d2d31946e87ac7
+size 123792