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": 76428.60421528526,
-      "required_pages_per_second_upper_estimate": 0.005287508178407732
+      "required_token_positions_per_second": 76872.58514691339,
+      "required_pages_per_second_upper_estimate": 0.005318223809435458
     },
     {
       "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": 9553.575526910658,
-      "required_pages_per_second_upper_estimate": 0.003999882571774437
+      "required_token_positions_per_second": 9609.073143364174,
+      "required_pages_per_second_upper_estimate": 0.004023118264861507
     },
     {
       "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": 19107.151053821315,
-      "required_pages_per_second_upper_estimate": 0.003979515675052592
+      "required_token_positions_per_second": 19218.146286728348,
+      "required_pages_per_second_upper_estimate": 0.004002633054925994
     },
     {
       "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": 28660.726580731975,
-      "required_pages_per_second_upper_estimate": 0.001718166796119318
+      "required_token_positions_per_second": 28827.219430092522,
+      "required_pages_per_second_upper_estimate": 0.0017281477882186216
     },
     {
       "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": 28660.726580731975,
-      "required_pages_per_second_upper_estimate": 0.002228245426438665
+      "required_token_positions_per_second": 28827.219430092522,
+      "required_pages_per_second_upper_estimate": 0.0022411895131517977
     },
     {
       "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": 28660.726580731975,
-      "required_pages_per_second_upper_estimate": 0.3676290270870304
+      "required_token_positions_per_second": 28827.219430092522,
+      "required_pages_per_second_upper_estimate": 0.36976461859253373
     }
   ],
   "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": 0.06535290633589952,
-    "producer_p95_seconds": 0.13868397399983223,
-    "gpu_update_seconds": 0.9646650067338903,
-    "producer_batches_per_second": 15.301538310480343,
-    "gpu_batches_per_second": 1.036629288944299,
-    "producer_headroom": 14.7608585573184,
+    "producer_batch_seconds": 0.2129036166015439,
+    "producer_p95_seconds": 0.1417908919997899,
+    "gpu_update_seconds": 0.9590935423739987,
+    "producer_batches_per_second": 4.696961075450084,
+    "gpu_batches_per_second": 1.0426511657296196,
+    "producer_headroom": 4.504825036246208,
     "sustainable": true,
-    "buffer_seconds": 60.77389542423509,
+    "buffer_seconds": 60.42289316956192,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 434867.1336098362,
-    "upload_bytes_per_second": 4112029.9246091954,
+    "generation_bytes_per_second": 483331.6186730267,
+    "upload_bytes_per_second": 4310434.583952965,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.31236368390384,
+    "estimated_full_shards_per_hour": 103.71171398299313,
     "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": [
@@ -292,6 +292,12 @@
       "bytes": 2456869,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.450166357999933,
+      "bytes": 12214165,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -24,3 +24,4 @@
 {"step": 442600, "loss": 0.5788451433181763, "lr": 0.0003, "grad_norm": 0.15982561989572341, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64783343283, "tokens_per_second": 153029.0152602861, "data_wait_s": 1.477799969507032e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 442700, "loss": 0.5815114378929138, "lr": 0.0003, "grad_norm": 0.15964106845182402, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64798010151, "tokens_per_second": 152454.13909953978, "data_wait_s": 1.5870999959588517e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 442800, "loss": 0.5626919269561768, "lr": 0.0003, "grad_norm": 0.16894435588564782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64812683395, "tokens_per_second": 153366.86244665916, "data_wait_s": 1.3889999991079094e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 442900, "loss": 0.5983905792236328, "lr": 0.0003, "grad_norm": 0.17057465081739728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64827354536, "tokens_per_second": 152355.02874252218, "data_wait_s": 1.4829000065219589e-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 sha256:2329693d470b6becf3fce62bd73b21ffeb4b562ea0c9b319c79455c52f90b027
-size 55233
+oid sha256:aa70234dcb48de8d56ef4c61687dd81b914b82b08ced445d6866ed08f61e9d51
+size 57351