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": 75525.28617801439,
-      "required_pages_per_second_upper_estimate": 0.006546340388822489
+      "required_token_positions_per_second": 75563.8333989353,
+      "required_pages_per_second_upper_estimate": 0.006549681564234875
     },
     {
       "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": 9440.660772251798,
-      "required_pages_per_second_upper_estimate": 0.0037660207325082968
+      "required_token_positions_per_second": 9445.479174866912,
+      "required_pages_per_second_upper_estimate": 0.003767942865353005
     },
     {
       "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": 18881.321544503597,
-      "required_pages_per_second_upper_estimate": 0.004367861439411175
+      "required_token_positions_per_second": 18890.958349733824,
+      "required_pages_per_second_upper_estimate": 0.004370090744699209
     },
     {
       "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": 28321.98231675539,
-      "required_pages_per_second_upper_estimate": 0.0017246601291331335
+      "required_token_positions_per_second": 28336.437524600737,
+      "required_pages_per_second_upper_estimate": 0.0017255403754503006
     },
     {
       "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": 28321.98231675539,
-      "required_pages_per_second_upper_estimate": 0.002180304599008723
+      "required_token_positions_per_second": 28336.437524600737,
+      "required_pages_per_second_upper_estimate": 0.002181417400923234
     },
     {
       "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": 28321.98231675539,
-      "required_pages_per_second_upper_estimate": 0.35166423279679393
+      "required_token_positions_per_second": 28336.437524600737,
+      "required_pages_per_second_upper_estimate": 0.35184371809976456
     }
   ],
   "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": 0.07979574096870579,
-    "producer_p95_seconds": 0.2057938079997257,
-    "gpu_update_seconds": 0.9762028551102985,
-    "producer_batches_per_second": 12.531997170026644,
-    "gpu_batches_per_second": 1.0243772539335718,
-    "producer_headroom": 12.23377141761419,
+    "producer_batch_seconds": 0.2758565923982985,
+    "producer_p95_seconds": 0.25630392899984145,
+    "gpu_update_seconds": 0.9757048667813991,
+    "producer_batches_per_second": 3.6250719669448364,
+    "gpu_batches_per_second": 1.024900084078441,
+    "producer_headroom": 3.537000360580896,
     "sustainable": true,
-    "buffer_seconds": 60.524577016838506,
+    "buffer_seconds": 60.493701740446745,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 167,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 593899.1755777331,
-    "upload_bytes_per_second": 5558565.909025971,
+    "generation_bytes_per_second": 546469.9338586339,
+    "upload_bytes_per_second": 5568822.611128862,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.71846890687463,
+    "estimated_full_shards_per_hour": 117.25972663707029,
     "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": 1.2188318490007077,
-      "bytes": 7592648,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.8659798669996235,
       "bytes": 5051955,
@@ -562,6 +556,12 @@
       "bytes": 5090460,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.1308283350008423,
+      "bytes": 7872807,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -93,3 +93,4 @@
 {"step": 289500, "loss": 0.5666735358536243, "lr": 0.0003, "grad_norm": 0.15751036352915404, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42374777933, "tokens_per_second": 150442.45961372167, "data_wait_s": 1.244700251845643e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 289600, "loss": 0.618221141397953, "lr": 0.0003, "grad_norm": 0.16659526595071078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42389317585, "tokens_per_second": 149192.4680953899, "data_wait_s": 6.915000994922593e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 289700, "loss": 0.6057268213480711, "lr": 0.0003, "grad_norm": 0.14388633062038758, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42403956512, "tokens_per_second": 150439.72960118204, "data_wait_s": 1.4343000657390803e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 289800, "loss": 0.5929397474974394, "lr": 0.0003, "grad_norm": 0.15009253358194619, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42418609542, "tokens_per_second": 150659.41350640537, "data_wait_s": 1.4784000086365268e-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:c9084c4a4dad1f99ba3b463edd4f7a2e0e5da1bb8050ac23b3d7ee8271e85c1e
-size 202158
+oid sha256:396ff63a7a363a279d698728b2111a13515b9a3b358b1b86b08575dc8aab56e1
+size 204276