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": 74656.33136042979,
-      "required_pages_per_second_upper_estimate": 0.0064710215875801065
+      "required_token_positions_per_second": 74203.99415940266,
+      "required_pages_per_second_upper_estimate": 0.00643181414543323
     },
     {
       "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": 9332.041420053723,
-      "required_pages_per_second_upper_estimate": 0.0037226908489124476
+      "required_token_positions_per_second": 9275.499269925333,
+      "required_pages_per_second_upper_estimate": 0.0037001353398457525
     },
     {
       "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": 18664.082840107447,
-      "required_pages_per_second_upper_estimate": 0.004317607088419742
+      "required_token_positions_per_second": 18550.998539850665,
+      "required_pages_per_second_upper_estimate": 0.004291447025771046
     },
     {
       "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": 27996.124260161174,
-      "required_pages_per_second_upper_estimate": 0.0017048170831316402
+      "required_token_positions_per_second": 27826.497809776,
+      "required_pages_per_second_upper_estimate": 0.0016944877222643884
     },
     {
       "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": 27996.124260161174,
-      "required_pages_per_second_upper_estimate": 0.002155219143779267
+      "required_token_positions_per_second": 27826.497809776,
+      "required_pages_per_second_upper_estimate": 0.0021421608300725486
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27996.124260161174,
-      "required_pages_per_second_upper_estimate": 0.3476181663106544
+      "required_token_positions_per_second": 27826.497809776,
+      "required_pages_per_second_upper_estimate": 0.3455119735000807
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2971332069533048,
-    "producer_p95_seconds": 0.34015339799952926,
-    "gpu_update_seconds": 0.9875652695020876,
-    "producer_batches_per_second": 3.365493915182467,
-    "gpu_batches_per_second": 1.0125912999190239,
-    "producer_headroom": 3.3236449053548087,
+    "producer_batch_seconds": 0.1007958861094096,
+    "producer_p95_seconds": 0.34759163900162093,
+    "gpu_update_seconds": 0.9935853296740315,
+    "producer_batches_per_second": 9.921039822146541,
+    "gpu_batches_per_second": 1.006456083976273,
+    "producer_headroom": 9.857399622396667,
     "sustainable": true,
-    "buffer_seconds": 60.24148143962734,
+    "buffer_seconds": 60.60870511011592,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 396704.5806527699,
-    "upload_bytes_per_second": 2800150.413639608,
+    "generation_bytes_per_second": 399262.56491846504,
+    "upload_bytes_per_second": 2755324.139936138,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.12356819808313,
+    "estimated_full_shards_per_hour": 85.6724520746752,
     "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": 1.268800834997819,
-      "bytes": 6327049,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.941673843000899,
       "bytes": 1008328,
@@ -562,6 +556,12 @@
       "bytes": 6662663,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 3.650391375998879,
+      "bytes": 6163332,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -253,3 +253,4 @@
 {"step": 220300, "loss": 0.5912884678691626, "lr": 0.0003, "grad_norm": 0.13969897268591489, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32245296649, "tokens_per_second": 149061.06156735765, "data_wait_s": 8.604998583905399e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 220400, "loss": 0.6047754622995853, "lr": 0.0003, "grad_norm": 0.16461691713491167, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32259873420, "tokens_per_second": 147838.31138350404, "data_wait_s": 8.69299765327014e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 220500, "loss": 0.5657390430569649, "lr": 0.0003, "grad_norm": 0.13903210897409343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32274366113, "tokens_per_second": 147752.0398607778, "data_wait_s": 1.3315999240148813e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 220600, "loss": 0.6197793558239937, "lr": 0.0003, "grad_norm": 0.15691857107383397, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32289077243, "tokens_per_second": 149457.04322871962, "data_wait_s": 1.5943998732836917e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cb787ee3d9935193b30d86dae6b497d2245ba0b83bbd9e6a7ebcdc6eb26d9b43
-size 544170
+oid sha256:f3ebc6f92ed051417e90be2d52cc5f0d4e67c76b10eb02f0b13a05e8224ba1d8
+size 546288