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": 75683.6698930679,
-      "required_pages_per_second_upper_estimate": 0.005235971893510968
+      "required_token_positions_per_second": 76453.60105814453,
+      "required_pages_per_second_upper_estimate": 0.005289237517997402
     },
     {
       "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": 9460.458736633487,
-      "required_pages_per_second_upper_estimate": 0.003960896516185083
+      "required_token_positions_per_second": 9556.700132268066,
+      "required_pages_per_second_upper_estimate": 0.0040011907787884035
     },
     {
       "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": 18920.917473266974,
-      "required_pages_per_second_upper_estimate": 0.003940728131532913
+      "required_token_positions_per_second": 19113.400264536132,
+      "required_pages_per_second_upper_estimate": 0.003980817220841719
     },
     {
       "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": 28381.376209900463,
-      "required_pages_per_second_upper_estimate": 0.0017014201679312875
+      "required_token_positions_per_second": 28670.1003968042,
+      "required_pages_per_second_upper_estimate": 0.0017187287420798595
     },
     {
       "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": 28381.376209900463,
-      "required_pages_per_second_upper_estimate": 0.0022065271638389396
+      "required_token_positions_per_second": 28670.1003968042,
+      "required_pages_per_second_upper_estimate": 0.0022289741994072207
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28381.376209900463,
-      "required_pages_per_second_upper_estimate": 0.364045820473063
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28670.1003968042,
+      "required_pages_per_second_upper_estimate": 0.36774926433478533
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09619037810924169,
+    "producer_batch_seconds": 0.12654482389845612,
     "producer_p95_seconds": 0.2847857099986868,
-    "gpu_update_seconds": 0.9741599489581962,
-    "producer_batches_per_second": 10.396050204359504,
-    "gpu_batches_per_second": 1.026525470554849,
-    "producer_headroom": 10.1274157364457,
+    "gpu_update_seconds": 0.9643496052452565,
+    "producer_batches_per_second": 7.902338232359738,
+    "gpu_batches_per_second": 1.0369683303242259,
+    "producer_headroom": 7.620616754890611,
     "sustainable": true,
-    "buffer_seconds": 60.39791683540817,
+    "buffer_seconds": 60.75402513045116,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 368348.9713090974,
-    "upload_bytes_per_second": 5868109.901260721,
+    "generation_bytes_per_second": 354955.35036680783,
+    "upload_bytes_per_second": 5937659.932236937,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.0391145177335,
+    "estimated_full_shards_per_hour": 76.16515525105645,
     "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": 1.0805805290001445,
-      "bytes": 1076352,
-      "pages": 29,
-      "shards": 1
-    },
     {
       "seconds": 1.1792509940023592,
       "bytes": 9839539,
@@ -562,6 +556,12 @@
       "bytes": 16684165,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.2760817240014148,
+      "bytes": 7386323,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -276,3 +276,4 @@
 {"step": 584700, "loss": 0.590989351272583, "lr": 0.0003, "grad_norm": 0.1806887773260374, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85582950419, "tokens_per_second": 152748.95279501256, "data_wait_s": 9.019000572152436e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 584800, "loss": 0.5795027613639832, "lr": 0.0003, "grad_norm": 0.20328860796973394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85597670119, "tokens_per_second": 142973.37827009548, "data_wait_s": 1.3257002137834206e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 584900, "loss": 0.5780168175697327, "lr": 0.0003, "grad_norm": 0.17205930086431406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85612460103, "tokens_per_second": 153380.65364789363, "data_wait_s": 7.360999006778002e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 585000, "loss": 0.6551768779754639, "lr": 0.0003, "grad_norm": 0.15627566228715561, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85627188784, "tokens_per_second": 152485.23323052283, "data_wait_s": 1.150599928223528e-05, "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:a232c36190975a2992e3dbd8355c231755f22b15d12163a8994e2efdc8487e4c
-size 592884
+oid sha256:7e359187a3c3ffed206bbb95de099fc2da50d2f1c6de3ff2f099027c238ebf0e
+size 595002