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": 77595.64990235234,
-      "required_pages_per_second_upper_estimate": 0.005368247106958104
+      "required_token_positions_per_second": 77065.30860704908,
+      "required_pages_per_second_upper_estimate": 0.005331556865587682
     },
     {
       "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": 9699.456237794042,
-      "required_pages_per_second_upper_estimate": 0.004060959779085656
+      "required_token_positions_per_second": 9633.163575881135,
+      "required_pages_per_second_upper_estimate": 0.004033204425890922
     },
     {
       "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": 19398.912475588084,
-      "required_pages_per_second_upper_estimate": 0.004040281884940501
+      "required_token_positions_per_second": 19266.32715176227,
+      "required_pages_per_second_upper_estimate": 0.004012667858497704
     },
     {
       "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": 29098.368713382126,
-      "required_pages_per_second_upper_estimate": 0.0017444027737308504
+      "required_token_positions_per_second": 28899.490727643406,
+      "required_pages_per_second_upper_estimate": 0.0017324803421549153
     },
     {
       "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": 29098.368713382126,
-      "required_pages_per_second_upper_estimate": 0.002262270177267911
+      "required_token_positions_per_second": 28899.490727643406,
+      "required_pages_per_second_upper_estimate": 0.0022468082886485364
     },
     {
       "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": 29098.368713382126,
-      "required_pages_per_second_upper_estimate": 0.37324263046115524
+      "required_token_positions_per_second": 28899.490727643406,
+      "required_pages_per_second_upper_estimate": 0.370691637198643
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07380505301561868,
-    "producer_p95_seconds": 0.1825122199998077,
-    "gpu_update_seconds": 0.9501563566099459,
-    "producer_batches_per_second": 13.549207800018506,
-    "gpu_batches_per_second": 1.0524583591356382,
-    "producer_headroom": 12.873865918216644,
+    "producer_batch_seconds": 0.10176490800007088,
+    "producer_p95_seconds": 0.1957815309997386,
+    "gpu_update_seconds": 0.9566950594583901,
+    "producer_batches_per_second": 9.82657007855108,
+    "gpu_batches_per_second": 1.0452651449523802,
+    "producer_headroom": 9.401031045571463,
     "sustainable": true,
-    "buffer_seconds": 60.81000682303654,
+    "buffer_seconds": 60.271788745878574,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 341520.4034042332,
-    "upload_bytes_per_second": 5579516.422005196,
+    "generation_bytes_per_second": 312609.2211451523,
+    "upload_bytes_per_second": 5467768.376693739,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.28232838244674,
+    "estimated_full_shards_per_hour": 67.07866168752601,
     "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": [
@@ -526,6 +526,12 @@
       "bytes": 6299752,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 3.5441821820004407,
+      "bytes": 11982351,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -68,3 +68,4 @@
 {"step": 563900, "loss": 0.5917801856994629, "lr": 0.0003, "grad_norm": 0.16982393404048088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82537652335, "tokens_per_second": 153940.50013117734, "data_wait_s": 1.5874999917286914e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 564000, "loss": 0.58759605884552, "lr": 0.0003, "grad_norm": 0.1551288773528406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82552269500, "tokens_per_second": 152797.98371624784, "data_wait_s": 9.035000402946025e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 564100, "loss": 0.5673473477363586, "lr": 0.0003, "grad_norm": 0.16710315307229787, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82566814893, "tokens_per_second": 148554.20675057062, "data_wait_s": 1.6445999790448695e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 564200, "loss": 0.6458748579025269, "lr": 0.0003, "grad_norm": 0.159881705543696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82581368181, "tokens_per_second": 151678.36277086398, "data_wait_s": 1.4647000170953106e-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 sha256:85bc499fdd13e67ad31bbe54048e62fcf1339193a52f7f572ba71bbe5e4495bd
-size 149208
+oid sha256:ed2c3be20eb76c45d7a2f1ed953207c1718176b8b520672d72043ccbb6c9c8d7
+size 151326