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": 77143.75945655801,
-      "required_pages_per_second_upper_estimate": 0.005336984277387759
+      "required_token_positions_per_second": 77032.11897381807,
+      "required_pages_per_second_upper_estimate": 0.005329260729750201
     },
     {
       "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": 9642.969932069751,
-      "required_pages_per_second_upper_estimate": 0.004037310142447092
+      "required_token_positions_per_second": 9629.014871727259,
+      "required_pages_per_second_upper_estimate": 0.004031467450096488
     },
     {
       "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": 19285.939864139502,
-      "required_pages_per_second_upper_estimate": 0.004016752669263874
+      "required_token_positions_per_second": 19258.029743454517,
+      "required_pages_per_second_upper_estimate": 0.004010939727164571
     },
     {
       "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": 28928.909796209253,
-      "required_pages_per_second_upper_estimate": 0.0017342439703951211
+      "required_token_positions_per_second": 28887.044615181778,
+      "required_pages_per_second_upper_estimate": 0.001731734216716954
     },
     {
       "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": 28928.909796209253,
-      "required_pages_per_second_upper_estimate": 0.0022490954918287236
+      "required_token_positions_per_second": 28887.044615181778,
+      "required_pages_per_second_upper_estimate": 0.002245840658149308
     },
     {
       "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": 28928.909796209253,
-      "required_pages_per_second_upper_estimate": 0.3710689934224709
+      "required_token_positions_per_second": 28887.044615181778,
+      "required_pages_per_second_upper_estimate": 0.3705319918315796
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.0877402217888914,
-    "producer_p95_seconds": 0.2536444430006668,
-    "gpu_update_seconds": 0.955722154577111,
-    "producer_batches_per_second": 11.397281424773054,
-    "gpu_batches_per_second": 1.0463292026985407,
-    "producer_headroom": 10.892634359605788,
+    "producer_batch_seconds": 0.627025644757623,
+    "producer_p95_seconds": 0.2556276089999301,
+    "gpu_update_seconds": 0.9571072558066189,
+    "producer_batches_per_second": 1.5948311019823604,
+    "gpu_batches_per_second": 1.044814981741239,
+    "producer_headroom": 1.5264244194933831,
     "sustainable": true,
-    "buffer_seconds": 60.21049573835799,
+    "buffer_seconds": 60.29775711581699,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 341665.40568189946,
-    "upload_bytes_per_second": 5738258.984628578,
+    "generation_bytes_per_second": 299137.4796276202,
+    "upload_bytes_per_second": 5707589.346256835,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.31344249575365,
+    "estimated_full_shards_per_hour": 64.18793956395582,
     "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": 1.010654464000254,
-      "bytes": 6096042,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.1592712370002118,
       "bytes": 7445713,
@@ -562,6 +556,12 @@
       "bytes": 7596077,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.3744402519987489,
+      "bytes": 5869308,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -219,3 +219,4 @@
 {"step": 706900, "loss": 0.5981852412223816, "lr": 8e-05, "grad_norm": 0.22925572127747484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103470381310, "tokens_per_second": 153123.57898828003, "data_wait_s": 1.2748001608997583e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 707000, "loss": 0.5401789546012878, "lr": 8e-05, "grad_norm": 0.22340848497574609, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103485085267, "tokens_per_second": 153346.46700654848, "data_wait_s": 1.2869000784121454e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 707100, "loss": 0.5317685008049011, "lr": 8e-05, "grad_norm": 0.21699219140543216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103499593639, "tokens_per_second": 149133.01091071288, "data_wait_s": 1.6828002117108554e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 707200, "loss": 0.5109959244728088, "lr": 8e-05, "grad_norm": 0.20483624901390615, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103514080865, "tokens_per_second": 150895.4630753194, "data_wait_s": 1.7610000213608146e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:760bce4f4047f960fcb9398025ec22a46cb1b0037e3beefdec0c7a02d58cc177
-size 471375
+oid sha256:689d06b6dcb9f77419b9447213dc4b09ac3b7e1928a5ee4959f888a633d659d4
+size 473493