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": 77200.54833695684,
-      "required_pages_per_second_upper_estimate": 0.005340913063902107
+      "required_token_positions_per_second": 76736.58371776751,
+      "required_pages_per_second_upper_estimate": 0.0053088149150003535
     },
     {
       "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": 9650.068542119605,
-      "required_pages_per_second_upper_estimate": 0.004040282182239132
+      "required_token_positions_per_second": 9592.07296472094,
+      "required_pages_per_second_upper_estimate": 0.004016000645067683
     },
     {
       "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": 19300.13708423921,
-      "required_pages_per_second_upper_estimate": 0.004019709575804772
+      "required_token_positions_per_second": 19184.14592944188,
+      "required_pages_per_second_upper_estimate": 0.0039955516771529405
     },
     {
       "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": 28950.205626358817,
-      "required_pages_per_second_upper_estimate": 0.0017355206228957375
+      "required_token_positions_per_second": 28776.21889416282,
+      "required_pages_per_second_upper_estimate": 0.0017250903839629453
     },
     {
       "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": 28950.205626358817,
-      "required_pages_per_second_upper_estimate": 0.0022507511489523947
+      "required_token_positions_per_second": 28776.21889416282,
+      "required_pages_per_second_upper_estimate": 0.002237224445810913
     },
     {
       "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": 28950.205626358817,
-      "required_pages_per_second_upper_estimate": 0.37134215346594857
+      "required_token_positions_per_second": 28776.21889416282,
+      "required_pages_per_second_upper_estimate": 0.36911043847773656
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1191355082657708,
-    "producer_p95_seconds": 0.25448327100093593,
-    "gpu_update_seconds": 0.955019123416064,
-    "producer_batches_per_second": 8.393803111740391,
-    "gpu_batches_per_second": 1.047099451184853,
-    "producer_headroom": 8.01624248990134,
+    "producer_batch_seconds": 0.4707425980782318,
+    "producer_p95_seconds": 0.25531490499997744,
+    "gpu_update_seconds": 0.9607933586301821,
+    "producer_batches_per_second": 2.1243031841231668,
+    "gpu_batches_per_second": 1.0408065282900325,
+    "producer_headroom": 2.0410163910224877,
     "sustainable": true,
-    "buffer_seconds": 60.16620477521203,
+    "buffer_seconds": 60.52998159370147,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374633.74285094044,
-    "upload_bytes_per_second": 5322022.455701511,
+    "generation_bytes_per_second": 320576.35359510063,
+    "upload_bytes_per_second": 5438002.823820035,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.38768018861923,
+    "estimated_full_shards_per_hour": 68.78822284593357,
     "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": [
-    {
-      "seconds": 1.031140414001129,
-      "bytes": 2762787,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.0752210870014096,
       "bytes": 5610482,
@@ -562,6 +556,12 @@
       "bytes": 1101958,
       "pages": 31,
       "shards": 1
+    },
+    {
+      "seconds": 1.268539120999776,
+      "bytes": 13168093,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -244,3 +244,4 @@
 {"step": 581500, "loss": 0.5811357498168945, "lr": 0.0003, "grad_norm": 0.14319670271985702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85114845007, "tokens_per_second": 153675.46980002642, "data_wait_s": 9.942999895429239e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 581600, "loss": 0.6176038384437561, "lr": 0.0003, "grad_norm": 0.16526034513122184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85129605826, "tokens_per_second": 153454.87485712368, "data_wait_s": 1.1123996955575421e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 581700, "loss": 0.578392505645752, "lr": 0.0003, "grad_norm": 0.17558227191054976, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85144395957, "tokens_per_second": 154703.03630238463, "data_wait_s": 1.0376999853178859e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 581800, "loss": 0.5910410284996033, "lr": 0.0003, "grad_norm": 0.19374622682089612, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85158879530, "tokens_per_second": 150689.796300987, "data_wait_s": 8.909999451134354e-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:dfa2e64a9aec6b52ad85dc0c2ff5a14e10be982717a427e5e30320e5a4ded7ec
-size 525108
+oid sha256:2d57d8950ebe076837d1778267dd7fd5dbbf200514f756460b2f3e6da70d7bc2
+size 527226