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": 75992.82542611643,
-      "required_pages_per_second_upper_estimate": 0.006586865505880583
+      "required_token_positions_per_second": 75864.0139628396,
+      "required_pages_per_second_upper_estimate": 0.0065757004547134165
     },
     {
       "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": 9499.103178264553,
-      "required_pages_per_second_upper_estimate": 0.003789334282058622
+      "required_token_positions_per_second": 9483.00174535495,
+      "required_pages_per_second_upper_estimate": 0.003782911179733106
     },
     {
       "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": 18998.206356529106,
-      "required_pages_per_second_upper_estimate": 0.004394900683570844
+      "required_token_positions_per_second": 18966.0034907099,
+      "required_pages_per_second_upper_estimate": 0.004387451117314652
     },
     {
       "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": 28497.309534793658,
-      "required_pages_per_second_upper_estimate": 0.001735336636840842
+      "required_token_positions_per_second": 28449.005236064848,
+      "required_pages_per_second_upper_estimate": 0.0017323951584813267
     },
     {
       "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": 28497.309534793658,
-      "required_pages_per_second_upper_estimate": 0.0021938017735901112
+      "required_token_positions_per_second": 28449.005236064848,
+      "required_pages_per_second_upper_estimate": 0.0021900831749591104
     },
     {
       "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": 28497.309534793658,
-      "required_pages_per_second_upper_estimate": 0.353841210062377
+      "required_token_positions_per_second": 28449.005236064848,
+      "required_pages_per_second_upper_estimate": 0.3532414323362535
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.1355042502342485,
-    "producer_p95_seconds": 0.4894277540006442,
-    "gpu_update_seconds": 0.9701968519604738,
-    "producer_batches_per_second": 7.379842316910968,
-    "gpu_batches_per_second": 1.0307186608359975,
-    "producer_headroom": 7.15989978383171,
+    "producer_batch_seconds": 0.11744457445325907,
+    "producer_p95_seconds": 0.3774816959994496,
+    "gpu_update_seconds": 0.9718441741840094,
+    "producer_batches_per_second": 8.514654718238882,
+    "gpu_batches_per_second": 1.0289715435497993,
+    "producer_headroom": 8.274917583108845,
     "sustainable": true,
-    "buffer_seconds": 60.152204821549375,
+    "buffer_seconds": 60.25433879940858,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 256798.19968139278,
-    "upload_bytes_per_second": 5261212.351291741,
+    "generation_bytes_per_second": 268006.0518125485,
+    "upload_bytes_per_second": 5166674.48614815,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.10291569548929,
+    "estimated_full_shards_per_hour": 57.50785985739079,
     "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": 0.9819577199996274,
-      "bytes": 10903211,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 0.860417497002345,
       "bytes": 864723,
@@ -562,6 +556,12 @@
       "bytes": 366522,
       "pages": 10,
       "shards": 1
+    },
+    {
+      "seconds": 0.8059965610009385,
+      "bytes": 2452391,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -350,3 +350,4 @@
 {"step": 390000, "loss": 0.6361445300281048, "lr": 0.0003, "grad_norm": 0.16592915181678689, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57085381676, "tokens_per_second": 140165.60097290107, "data_wait_s": 1.0950003343168646e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 390100, "loss": 0.5821457048878074, "lr": 0.0003, "grad_norm": 0.15114542171762976, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57100091088, "tokens_per_second": 132205.08004367616, "data_wait_s": 8.513001375831664e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 390200, "loss": 0.5980184301733971, "lr": 0.0003, "grad_norm": 0.16292192317862475, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57114717086, "tokens_per_second": 148854.8882911339, "data_wait_s": 8.458999218419194e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 390300, "loss": 0.6017175558954477, "lr": 0.0003, "grad_norm": 0.1479221209816008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57129537823, "tokens_per_second": 151049.98714732862, "data_wait_s": 8.486000297125429e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3fb0dd554c052c460071594af63c667fb9754e0c33e85a64fb9f244b41c3f2e1
-size 751182
+oid sha256:a9c96ee4714e0e67d760aa2f8cf25bb1e077d9013898a711c6a3e2aaa31ae362
+size 753300