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": 76835.0369085418,
-      "required_pages_per_second_upper_estimate": 0.005315626135180995
+      "required_token_positions_per_second": 76375.08464873706,
+      "required_pages_per_second_upper_estimate": 0.0052838055706114235
     },
     {
       "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": 9604.379613567726,
-      "required_pages_per_second_upper_estimate": 0.004021153181947781
+      "required_token_positions_per_second": 9546.885581092132,
+      "required_pages_per_second_upper_estimate": 0.003997081631162175
     },
     {
       "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": 19208.75922713545,
-      "required_pages_per_second_upper_estimate": 0.004000677977966202
+      "required_token_positions_per_second": 19093.771162184265,
+      "required_pages_per_second_upper_estimate": 0.003976728996476065
     },
     {
       "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": 28813.138840703177,
-      "required_pages_per_second_upper_estimate": 0.0017273036783845455
+      "required_token_positions_per_second": 28640.656743276395,
+      "required_pages_per_second_upper_estimate": 0.0017169636400087235
     },
     {
       "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": 28813.138840703177,
-      "required_pages_per_second_upper_estimate": 0.0022400948092607433
+      "required_token_positions_per_second": 28640.656743276395,
+      "required_pages_per_second_upper_estimate": 0.002226685084854379
     },
     {
       "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": 28813.138840703177,
-      "required_pages_per_second_upper_estimate": 0.3695840079104062
+      "required_token_positions_per_second": 28640.656743276395,
+      "required_pages_per_second_upper_estimate": 0.3673715927614627
     }
   ],
   "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": 0.09595001377311974,
-    "producer_p95_seconds": 0.3511579029946006,
-    "gpu_update_seconds": 0.9595622383543568,
-    "producer_batches_per_second": 10.422093345026164,
-    "gpu_batches_per_second": 1.042141885152748,
-    "producer_headroom": 10.000647218491352,
+    "producer_batch_seconds": 0.1464352092263539,
+    "producer_p95_seconds": 0.3603584169977694,
+    "gpu_update_seconds": 0.9653409922763231,
+    "producer_batches_per_second": 6.828958727093008,
+    "gpu_batches_per_second": 1.0359033833650317,
+    "producer_headroom": 6.5922737938260205,
     "sustainable": true,
-    "buffer_seconds": 60.45242101632448,
+    "buffer_seconds": 60.81648251340835,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 173,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 482776.51616512955,
+    "generation_bytes_per_second": 634800.4779717487,
     "upload_bytes_per_second": 5661927.297692053,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.59260190692343,
+    "estimated_full_shards_per_hour": 68.10670258695767,
     "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": [

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

@@ -405,3 +405,4 @@
 {"step": 597600, "loss": 0.6045230627059937, "lr": 0.0003, "grad_norm": 0.17142739197484846, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87470801221, "tokens_per_second": 153084.1947797676, "data_wait_s": 1.292899833060801e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 597700, "loss": 0.6102030873298645, "lr": 0.0003, "grad_norm": 0.16332613550061487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87485597805, "tokens_per_second": 152967.56411244668, "data_wait_s": 1.6800004232209176e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 597800, "loss": 0.5536242723464966, "lr": 0.0003, "grad_norm": 0.16810066288829437, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87500102667, "tokens_per_second": 149522.73817017005, "data_wait_s": 1.5345998690463603e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 597900, "loss": 0.5882132649421692, "lr": 0.0003, "grad_norm": 0.1690358589646184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87514559626, "tokens_per_second": 150463.82559511013, "data_wait_s": 1.3912998838350177e-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 sha256:6e30c5afab4fb8101576ecdb0573fc8cbdb5754a8ab35037019a52fdcf705496
-size 868455
+oid sha256:b1dd11b2543ae11c3948ef3b79c2f4508793ef7212e106d4dc7394b64fc356dd
+size 870573