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": 77253.10734282904,
-      "required_pages_per_second_upper_estimate": 0.005344549217882041
+      "required_token_positions_per_second": 77278.7753550024,
+      "required_pages_per_second_upper_estimate": 0.005346324990522189
     },
     {
       "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": 9656.63841785363,
-      "required_pages_per_second_upper_estimate": 0.004043032852014361
+      "required_token_positions_per_second": 9659.8469193753,
+      "required_pages_per_second_upper_estimate": 0.004044376184600355
     },
     {
       "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": 19313.27683570726,
-      "required_pages_per_second_upper_estimate": 0.004022446239517018
+      "required_token_positions_per_second": 19319.6938387506,
+      "required_pages_per_second_upper_estimate": 0.004023782732023196
     },
     {
       "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": 28969.91525356089,
-      "required_pages_per_second_upper_estimate": 0.0017367021849516438
+      "required_token_positions_per_second": 28979.540758125906,
+      "required_pages_per_second_upper_estimate": 0.001737279219253022
     },
     {
       "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": 28969.91525356089,
-      "required_pages_per_second_upper_estimate": 0.002252283485773868
+      "required_token_positions_per_second": 28979.540758125906,
+      "required_pages_per_second_upper_estimate": 0.002253031826415678
     },
     {
       "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": 28969.91525356089,
-      "required_pages_per_second_upper_estimate": 0.37159496740114784
+      "required_token_positions_per_second": 28979.540758125906,
+      "required_pages_per_second_upper_estimate": 0.3717184330386463
     }
   ],
   "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": 0.28509248830469147,
-    "producer_p95_seconds": 0.16099925300022733,
-    "gpu_update_seconds": 0.9543693779567269,
-    "producer_batches_per_second": 3.507633631270053,
-    "gpu_batches_per_second": 1.0478123283261316,
-    "producer_headroom": 3.3475781267752955,
+    "producer_batch_seconds": 0.09882122220310663,
+    "producer_p95_seconds": 0.16641999500006932,
+    "gpu_update_seconds": 0.9540523858110989,
+    "producer_batches_per_second": 10.119283871481638,
+    "gpu_batches_per_second": 1.0481604730224936,
+    "producer_headroom": 9.65432692028683,
     "sustainable": true,
-    "buffer_seconds": 60.1252708112738,
+    "buffer_seconds": 60.10530030609923,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 446939.7232812257,
-    "upload_bytes_per_second": 3891025.8798871096,
+    "generation_bytes_per_second": 446304.953017198,
+    "upload_bytes_per_second": 3942747.161220033,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.90286039187983,
+    "estimated_full_shards_per_hour": 95.76665346991497,
     "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": [
@@ -406,6 +406,12 @@
       "bytes": 6934469,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0243737840000904,
+      "bytes": 7482664,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -48,3 +48,4 @@
 {"step": 445000, "loss": 0.6301952004432678, "lr": 0.0003, "grad_norm": 0.20991914998299785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65134552525, "tokens_per_second": 152765.13751353952, "data_wait_s": 1.3879000107408501e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 445100, "loss": 0.6043710708618164, "lr": 0.0003, "grad_norm": 0.14340291641602193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65149310375, "tokens_per_second": 139345.12325619237, "data_wait_s": 1.451499974791659e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 445200, "loss": 0.6336721777915955, "lr": 0.0003, "grad_norm": 0.17824873338584174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65163980123, "tokens_per_second": 153330.52134395167, "data_wait_s": 1.343100029771449e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 445300, "loss": 0.5626630783081055, "lr": 0.0003, "grad_norm": 0.16750663043162037, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65178632207, "tokens_per_second": 152824.15368629096, "data_wait_s": 9.791000593395438e-06, "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:db559e386282607083703c063febc878f7b940e1348236d6bf9439cec10b93d0
-size 106848
+oid sha256:a1e129305c8a2a4ad1e91c8f2aed224332a2e1b68968f482bfe0920d5e776149
+size 108966