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": 75451.32410800246,
-      "required_pages_per_second_upper_estimate": 0.006539929544049008
+      "required_token_positions_per_second": 75482.6948035512,
+      "required_pages_per_second_upper_estimate": 0.006542648676430872
     },
     {
       "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": 9431.415513500307,
-      "required_pages_per_second_upper_estimate": 0.0037623326605633906
+      "required_token_positions_per_second": 9435.3368504439,
+      "required_pages_per_second_upper_estimate": 0.003763896940499402
     },
     {
       "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": 18862.831027000615,
-      "required_pages_per_second_upper_estimate": 0.004363583983608854
+      "required_token_positions_per_second": 18870.6737008878,
+      "required_pages_per_second_upper_estimate": 0.004365398248186308
     },
     {
       "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": 28294.246540500924,
-      "required_pages_per_second_upper_estimate": 0.001722971165877608
+      "required_token_positions_per_second": 28306.010551331703,
+      "required_pages_per_second_upper_estimate": 0.0017236875324162077
     },
     {
       "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": 28294.246540500924,
-      "required_pages_per_second_upper_estimate": 0.0021781694221751115
+      "required_token_positions_per_second": 28306.010551331703,
+      "required_pages_per_second_upper_estimate": 0.002179075048293731
     },
     {
       "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": 28294.246540500924,
-      "required_pages_per_second_upper_estimate": 0.35131984728138527
+      "required_token_positions_per_second": 28306.010551331703,
+      "required_pages_per_second_upper_estimate": 0.35146591692429197
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.4558768907110675,
-    "producer_p95_seconds": 0.33596465600203373,
-    "gpu_update_seconds": 0.9771597897270078,
-    "producer_batches_per_second": 2.1935746697759133,
-    "gpu_batches_per_second": 1.0233740791558494,
-    "producer_headroom": 2.1434729630687217,
+    "producer_batch_seconds": 0.09516802138273306,
+    "producer_p95_seconds": 0.3320435979985632,
+    "gpu_update_seconds": 0.9767536809845234,
+    "producer_batches_per_second": 10.507731331077524,
+    "gpu_batches_per_second": 1.0237995714457357,
+    "producer_headroom": 10.263465256426379,
     "sustainable": true,
-    "buffer_seconds": 60.58390696307448,
+    "buffer_seconds": 60.558728221040454,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 270993.8079416989,
-    "upload_bytes_per_second": 3848255.54102955,
+    "generation_bytes_per_second": 260357.8991536015,
+    "upload_bytes_per_second": 3921666.148779026,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.14896277130341,
+    "estimated_full_shards_per_hour": 55.86674433666261,
     "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": 0.945849695999641,
-      "bytes": 1129028,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.1423426910005219,
       "bytes": 5919036,
@@ -562,6 +556,12 @@
       "bytes": 13411793,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.3467935920016316,
+      "bytes": 10736953,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -261,3 +261,4 @@
 {"step": 141100, "loss": 0.590720796957612, "lr": 0.0003, "grad_norm": 0.1466954825012965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20649955142, "tokens_per_second": 148243.05961548552, "data_wait_s": 1.7220998415723443e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 141200, "loss": 0.6340356022119522, "lr": 0.0003, "grad_norm": 0.1577793717491402, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20664575783, "tokens_per_second": 133969.28370366272, "data_wait_s": 1.1880001693498343e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 141300, "loss": 0.608159989118576, "lr": 0.0003, "grad_norm": 0.13898237365434615, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20679099854, "tokens_per_second": 148536.26547485095, "data_wait_s": 1.4794000890105963e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 141400, "loss": 0.6273522954434156, "lr": 0.0003, "grad_norm": 0.13436361468768224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20693775993, "tokens_per_second": 149429.81973961982, "data_wait_s": 1.2149997928645462e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3a36557bfc35590e4118b5033f7dca789f9140b37bef64e5c9f49f33b5e8323a
-size 561114
+oid sha256:7303be48583dd0854d920805fd31234267e334297daed8a4636b82799f45c717
+size 563232