Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76019.64168623484,
-      "required_pages_per_second_upper_estimate": 0.006589189871342457
+      "required_token_positions_per_second": 75435.82989465859,
+      "required_pages_per_second_upper_estimate": 0.006538586544906085
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9502.455210779355,
-      "required_pages_per_second_upper_estimate": 0.0037906714579461287
+      "required_token_positions_per_second": 9429.478736832323,
+      "required_pages_per_second_upper_estimate": 0.00376156005139314
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19004.91042155871,
-      "required_pages_per_second_upper_estimate": 0.004396451551027928
+      "required_token_positions_per_second": 18858.957473664646,
+      "required_pages_per_second_upper_estimate": 0.0043626879052167875
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28507.365632338064,
-      "required_pages_per_second_upper_estimate": 0.0017359490004210303
+      "required_token_positions_per_second": 28288.43621049697,
+      "required_pages_per_second_upper_estimate": 0.001722617347264813
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28507.365632338064,
-      "required_pages_per_second_upper_estimate": 0.0021945759198161393
+      "required_token_positions_per_second": 28288.43621049697,
+      "required_pages_per_second_upper_estimate": 0.0021777221268873843
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28507.365632338064,
-      "required_pages_per_second_upper_estimate": 0.35396607313828504
+      "required_token_positions_per_second": 28288.43621049697,
+      "required_pages_per_second_upper_estimate": 0.3512477024286597
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.10932569764088385,
-    "producer_p95_seconds": 0.342188292001083,
-    "gpu_update_seconds": 0.9698546107900191,
-    "producer_batches_per_second": 9.146980276172838,
-    "gpu_batches_per_second": 1.0310823796418571,
-    "producer_headroom": 8.87124099565159,
+    "producer_batch_seconds": 0.14992564555484478,
+    "producer_p95_seconds": 0.3747693879995495,
+    "gpu_update_seconds": 0.9773604943825306,
+    "producer_batches_per_second": 6.669972947584787,
+    "gpu_batches_per_second": 1.023163925437535,
+    "producer_headroom": 6.518968057569572,
     "sustainable": true,
-    "buffer_seconds": 60.130985868981185,
+    "buffer_seconds": 60.5963506517169,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 170,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 456792.3777017991,
-    "upload_bytes_per_second": 5632662.05979758,
+    "generation_bytes_per_second": 606136.9209482613,
+    "upload_bytes_per_second": 5566184.2658326505,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.01701067247848,
+    "estimated_full_shards_per_hour": 65.03143654506626,
     "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": 1.242159831999743,
-      "bytes": 12337974,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 1.3698127219977323,
       "bytes": 6046909,
@@ -562,6 +556,12 @@
       "bytes": 798107,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 0.9836393130026408,
+      "bytes": 5980946,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -307,3 +307,4 @@
 {"step": 310900, "loss": 0.5864830911159515, "lr": 0.0003, "grad_norm": 0.1517254468936491, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45507580553, "tokens_per_second": 151239.32859975062, "data_wait_s": 8.03900184109807e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 311000, "loss": 0.5390106067061424, "lr": 0.0003, "grad_norm": 0.13726870435127175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45522397256, "tokens_per_second": 152254.52151105856, "data_wait_s": 1.030000566970557e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 311100, "loss": 0.5741534139961004, "lr": 0.0003, "grad_norm": 0.13278086682221193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45537060365, "tokens_per_second": 150512.8172377953, "data_wait_s": 7.806003850419074e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 311200, "loss": 0.6199047211557627, "lr": 0.0003, "grad_norm": 0.17804002496439653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45551524664, "tokens_per_second": 148861.33297630894, "data_wait_s": 1.0080999345518649e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:52156603e1008a04fa9379677677338c3c9941bfd773c4a650bcd3171235ce3e
-size 659325
+oid sha256:703cd6ccda67a0b9be8bb3455dfa76ffe685bdb5eef10e0aab42db1a671aff7a
+size 661443