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": 75952.4825342208,
-      "required_pages_per_second_upper_estimate": 0.006583368686259195
+      "required_token_positions_per_second": 75897.20693643304,
+      "required_pages_per_second_upper_estimate": 0.006578577537537674
     },
     {
       "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": 9494.0603167776,
-      "required_pages_per_second_upper_estimate": 0.0037873226092139784
+      "required_token_positions_per_second": 9487.15086705413,
+      "required_pages_per_second_upper_estimate": 0.003784566326413806
     },
     {
       "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": 18988.1206335552,
-      "required_pages_per_second_upper_estimate": 0.004392567529063493
+      "required_token_positions_per_second": 18974.30173410826,
+      "required_pages_per_second_upper_estimate": 0.004389370769880767
     },
     {
       "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": 28482.180950332804,
-      "required_pages_per_second_upper_estimate": 0.0017344153854207246
+      "required_token_positions_per_second": 28461.452601162393,
+      "required_pages_per_second_upper_estimate": 0.0017331531377094897
     },
     {
       "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": 28482.180950332804,
-      "required_pages_per_second_upper_estimate": 0.0021926371332797127
+      "required_token_positions_per_second": 28461.452601162393,
+      "required_pages_per_second_upper_estimate": 0.0021910414075808314
     },
     {
       "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": 28482.180950332804,
-      "required_pages_per_second_upper_estimate": 0.35365336367548833
+      "required_token_positions_per_second": 28461.452601162393,
+      "required_pages_per_second_upper_estimate": 0.3533959869521138
     }
   ],
   "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": 0.08103640543748725,
-    "producer_p95_seconds": 0.29974260100061656,
-    "gpu_update_seconds": 0.9707121813534066,
-    "producer_batches_per_second": 12.340132741591255,
-    "gpu_batches_per_second": 1.0301714753447917,
-    "producer_headroom": 11.97871717178064,
+    "producer_batch_seconds": 0.12217297470311905,
+    "producer_p95_seconds": 0.30251249499997357,
+    "gpu_update_seconds": 0.9714191467118172,
+    "producer_batches_per_second": 8.185116245470859,
+    "gpu_batches_per_second": 1.0294217520675055,
+    "producer_headroom": 7.951178638912334,
     "sustainable": true,
-    "buffer_seconds": 60.18415524391121,
+    "buffer_seconds": 60.227987096132665,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 159,
+    "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 563978.0897682094,
-    "upload_bytes_per_second": 4954328.8602289,
+    "generation_bytes_per_second": 747818.9439282079,
+    "upload_bytes_per_second": 4934285.742726189,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.508284663127476,
+    "estimated_full_shards_per_hour": 80.23226851646746,
     "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": 0.9424325320032949,
-      "bytes": 6190430,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.982011802996567,
       "bytes": 5926739,
@@ -562,6 +556,12 @@
       "bytes": 10683433,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.2693092910012638,
+      "bytes": 6119669,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -265,3 +265,4 @@
 {"step": 381500, "loss": 0.5936720445752144, "lr": 0.0003, "grad_norm": 0.15501056398168583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55840213513, "tokens_per_second": 149718.14063646214, "data_wait_s": 1.219599653268233e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 381600, "loss": 0.573487963527441, "lr": 0.0003, "grad_norm": 0.1656134459225898, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55854726840, "tokens_per_second": 147647.3880881195, "data_wait_s": 1.3763001334154978e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 381700, "loss": 0.613114608451724, "lr": 0.0003, "grad_norm": 0.15971321352573042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55869369087, "tokens_per_second": 149673.83336199875, "data_wait_s": 1.0146999557036906e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 381800, "loss": 0.6879843585193157, "lr": 0.0003, "grad_norm": 0.16391770986705295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55884028395, "tokens_per_second": 149425.7625419428, "data_wait_s": 1.1308999091852456e-05, "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:d98c12469fa1d707d2aebdee0c1c1cc8c616bcb9698cf6ae1abac8bd2533ed79
-size 569586
+oid sha256:f833ff28da4431318654f5c5d8ee2c75a60da7f3a7e02162778fa179b64d9ccc
+size 571704