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": 75642.83946166119,
-      "required_pages_per_second_upper_estimate": 0.00655652960951292
+      "required_token_positions_per_second": 75480.03960756774,
+      "required_pages_per_second_upper_estimate": 0.006542418530772558
     },
     {
       "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": 9455.354932707649,
-      "required_pages_per_second_upper_estimate": 0.0037718824528114123
+      "required_token_positions_per_second": 9435.004950945968,
+      "required_pages_per_second_upper_estimate": 0.0037637645408273366
     },
     {
       "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": 18910.709865415298,
-      "required_pages_per_second_upper_estimate": 0.004374659910237314
+      "required_token_positions_per_second": 18870.009901891935,
+      "required_pages_per_second_upper_estimate": 0.0043652446899710776
     },
     {
       "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": 28366.064798122945,
-      "required_pages_per_second_upper_estimate": 0.0017273445209655123
+      "required_token_positions_per_second": 28305.0148528379,
+      "required_pages_per_second_upper_estimate": 0.0017236268996019615
     },
     {
       "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": 28366.064798122945,
-      "required_pages_per_second_upper_estimate": 0.0021836981904525157
+      "required_token_positions_per_second": 28305.0148528379,
+      "required_pages_per_second_upper_estimate": 0.0021789983966674083
     },
     {
       "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": 28366.064798122945,
-      "required_pages_per_second_upper_estimate": 0.3522115896808044
+      "required_token_positions_per_second": 28305.0148528379,
+      "required_pages_per_second_upper_estimate": 0.3514535536813874
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07561832105474764,
-    "producer_p95_seconds": 0.21367903200007277,
-    "gpu_update_seconds": 0.9746857802365854,
-    "producer_batches_per_second": 13.224308422240695,
-    "gpu_batches_per_second": 1.0259716723858125,
-    "producer_headroom": 12.889545372620917,
+    "producer_batch_seconds": 0.17613392810159212,
+    "producer_p95_seconds": 0.2363980560003256,
+    "gpu_update_seconds": 0.9767880406968934,
+    "producer_batches_per_second": 5.677497860737035,
+    "gpu_batches_per_second": 1.0237635580453524,
+    "producer_headroom": 5.545712011450132,
     "sustainable": true,
-    "buffer_seconds": 60.430518374668296,
+    "buffer_seconds": 60.56085852320739,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 347763.882051382,
-    "upload_bytes_per_second": 5813751.438737849,
+    "generation_bytes_per_second": 446011.5730089509,
+    "upload_bytes_per_second": 5802890.672378316,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.62203355938048,
+    "estimated_full_shards_per_hour": 95.7037009496822,
     "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": [
@@ -352,6 +352,12 @@
       "bytes": 1134712,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.1013248220006062,
+      "bytes": 6047733,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -82,3 +82,4 @@
 {"step": 93200, "loss": 0.6181947682052851, "lr": 0.0003, "grad_norm": 0.15533395151872798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13639367228, "tokens_per_second": 149947.11374554937, "data_wait_s": 1.005100057227537e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 93300, "loss": 0.5518498849123716, "lr": 0.0003, "grad_norm": 0.14331970414926384, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13653941869, "tokens_per_second": 148786.2114419632, "data_wait_s": 1.0289000783814117e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 93400, "loss": 0.5855385158210993, "lr": 0.0003, "grad_norm": 0.14450430002881498, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13668456976, "tokens_per_second": 148402.31966318795, "data_wait_s": 1.343600160907954e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 93500, "loss": 0.6189540736377239, "lr": 0.0003, "grad_norm": 0.12821118007150253, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13682944329, "tokens_per_second": 148068.26681549716, "data_wait_s": 1.4943001588108018e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:80b9abcf37460aa66d3123a5d193fdfd6c1608ded0029597f433ac6c91d44adf
-size 178860
+oid sha256:7f694e52a73b3254fe469f8c85f934a87414e6866de338c76ea2e7cccfec8c36
+size 180978