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": 76095.2926312072,
-      "required_pages_per_second_upper_estimate": 0.006595747103517094
+      "required_token_positions_per_second": 75575.57063987092,
+      "required_pages_per_second_upper_estimate": 0.0065506989185314096
     },
     {
       "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": 9511.9115789009,
-      "required_pages_per_second_upper_estimate": 0.0037944437445751153
+      "required_token_positions_per_second": 9446.946329983864,
+      "required_pages_per_second_upper_estimate": 0.0037685281354650807
     },
     {
       "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": 19023.8231578018,
-      "required_pages_per_second_upper_estimate": 0.004400826679704977
+      "required_token_positions_per_second": 18893.89265996773,
+      "required_pages_per_second_upper_estimate": 0.004370769545729726
     },
     {
       "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": 28535.734736702696,
-      "required_pages_per_second_upper_estimate": 0.0017376765300356487
+      "required_token_positions_per_second": 28340.83898995159,
+      "required_pages_per_second_upper_estimate": 0.0017258084015974635
     },
     {
       "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": 28535.734736702696,
-      "required_pages_per_second_upper_estimate": 0.0021967598520008357
+      "required_token_positions_per_second": 28340.83898995159,
+      "required_pages_per_second_upper_estimate": 0.0021817562379099786
     },
     {
       "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": 28535.734736702696,
-      "required_pages_per_second_upper_estimate": 0.35431832246920913
+      "required_token_positions_per_second": 28340.83898995159,
+      "required_pages_per_second_upper_estimate": 0.3518983695686652
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.15591379182808396,
-    "producer_p95_seconds": 0.37562428999808617,
-    "gpu_update_seconds": 0.9688904195075485,
-    "producer_batches_per_second": 6.413800782310748,
-    "gpu_batches_per_second": 1.0321084612522677,
-    "producer_headroom": 6.214270130610903,
+    "producer_batch_seconds": 0.5512204409690753,
+    "producer_p95_seconds": 0.36694224899838446,
+    "gpu_update_seconds": 0.9755533352348094,
+    "producer_batches_per_second": 1.8141562352839202,
+    "gpu_batches_per_second": 1.0250592805972074,
+    "producer_headroom": 1.7698061659682538,
     "sustainable": true,
-    "buffer_seconds": 60.07120600946801,
+    "buffer_seconds": 60.48430678455818,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 268604.3660450082,
-    "upload_bytes_per_second": 5705606.930368792,
+    "generation_bytes_per_second": 247527.0354288347,
+    "upload_bytes_per_second": 5752473.533639712,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.636244163634146,
+    "estimated_full_shards_per_hour": 53.1135396685484,
     "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": 1.5244087760002003,
-      "bytes": 914665,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.1577779409999494,
       "bytes": 7413917,
@@ -562,6 +556,12 @@
       "bytes": 10539391,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 0.8672450950034545,
+      "bytes": 606235,
+      "pages": 21,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -281,3 +281,4 @@
 {"step": 308300, "loss": 0.5641783438622952, "lr": 0.0003, "grad_norm": 0.18553824204848238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45126962412, "tokens_per_second": 149157.4715018199, "data_wait_s": 9.802002750802785e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 308400, "loss": 0.5799893271178007, "lr": 0.0003, "grad_norm": 0.16008279980643045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45141484024, "tokens_per_second": 148829.09510379934, "data_wait_s": 1.3167998986318707e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 308500, "loss": 0.5653564743697643, "lr": 0.0003, "grad_norm": 0.13787916471794054, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45156077775, "tokens_per_second": 150522.99730350787, "data_wait_s": 1.743799657560885e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 308600, "loss": 0.5770511124283075, "lr": 0.0003, "grad_norm": 0.1516405700960314, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45170720467, "tokens_per_second": 131558.55300296092, "data_wait_s": 1.0547999409027398e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4e506fb675dfedf431604237c086e636c4119db49840fe60da41a60dfd75f277
-size 603474
+oid sha256:63da1f9c87f24c53750c0d670a6545882ed12b5fa661398e270573d59ad12fe4
+size 605592