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": 74511.58170452456,
-      "required_pages_per_second_upper_estimate": 0.006458475054270891
+      "required_token_positions_per_second": 73229.38303073635,
+      "required_pages_per_second_upper_estimate": 0.0063473373229297565
     },
     {
       "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": 9313.94771306557,
-      "required_pages_per_second_upper_estimate": 0.003715472998669846
+      "required_token_positions_per_second": 9153.672878842044,
+      "required_pages_per_second_upper_estimate": 0.0036515369709757637
     },
     {
       "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": 18627.89542613114,
-      "required_pages_per_second_upper_estimate": 0.004309235767073058
+      "required_token_positions_per_second": 18307.34575768409,
+      "required_pages_per_second_upper_estimate": 0.004235082242759316
     },
     {
       "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": 27941.843139196713,
-      "required_pages_per_second_upper_estimate": 0.0017015116476559364
+      "required_token_positions_per_second": 27461.018636526132,
+      "required_pages_per_second_upper_estimate": 0.0016722319581344996
     },
     {
       "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": 27941.843139196713,
-      "required_pages_per_second_upper_estimate": 0.002151040432827663
+      "required_token_positions_per_second": 27461.018636526132,
+      "required_pages_per_second_upper_estimate": 0.0021140252316046697
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27941.843139196713,
-      "required_pages_per_second_upper_estimate": 0.34694417645550135
+      "required_token_positions_per_second": 27461.018636526132,
+      "required_pages_per_second_upper_estimate": 0.34097394534842534
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12054120989080275,
-    "producer_p95_seconds": 0.4255806700020912,
-    "gpu_update_seconds": 0.9894837596169699,
-    "producer_batches_per_second": 8.29591805910934,
-    "gpu_batches_per_second": 1.0106280070600664,
-    "producer_headroom": 8.208676190601826,
+    "producer_batch_seconds": 0.1087942699060136,
+    "producer_p95_seconds": 0.40784374700160697,
+    "gpu_update_seconds": 1.0068089740569623,
+    "producer_batches_per_second": 9.191660561387021,
+    "gpu_batches_per_second": 0.9932370745271316,
+    "producer_headroom": 9.254246339689908,
     "sustainable": true,
-    "buffer_seconds": 60.358509336635166,
+    "buffer_seconds": 60.40853844341774,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 359104.02409992594,
+    "generation_bytes_per_second": 406368.9948772478,
     "upload_bytes_per_second": 5053663.424818994,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.05536405800184,
+    "estimated_full_shards_per_hour": 87.1973265146072,
     "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-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -414,3 +414,4 @@
 {"step": 41400, "loss": 0.6243662759661674, "lr": 0.0003, "grad_norm": 0.1266594176148852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6057938636, "tokens_per_second": 147845.282010544, "data_wait_s": 1.0493000445421785e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 41500, "loss": 0.6293739005923271, "lr": 0.0003, "grad_norm": 0.13255592951291514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6072666804, "tokens_per_second": 148697.67904327752, "data_wait_s": 1.361100294161588e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 41600, "loss": 0.6368713565170765, "lr": 0.0003, "grad_norm": 0.15394438290790816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6087361539, "tokens_per_second": 148919.23862411396, "data_wait_s": 1.1239004379604012e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 41700, "loss": 0.6451467759907246, "lr": 0.0003, "grad_norm": 0.1327683823621243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6102058721, "tokens_per_second": 148677.4958583484, "data_wait_s": 1.2882002920378e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:489ceeb471dad4ef0b05021bdaf3878674acd0139de0278a897a1a3781872280
-size 881743
+oid sha256:a35cff1974ce05b362a03eace6b69880fbbfeffadea8e0f1687537caf9b0f055
+size 883861