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": 75252.7795074329,
-      "required_pages_per_second_upper_estimate": 0.006522720201278328
+      "required_token_positions_per_second": 75302.43386295743,
+      "required_pages_per_second_upper_estimate": 0.006527024114967381
     },
     {
       "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": 9406.597438429113,
-      "required_pages_per_second_upper_estimate": 0.0037524323593542018
+      "required_token_positions_per_second": 9412.80423286968,
+      "required_pages_per_second_upper_estimate": 0.0037549083424563904
     },
     {
       "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": 18813.194876858226,
-      "required_pages_per_second_upper_estimate": 0.004352101533839818
+      "required_token_positions_per_second": 18825.60846573936,
+      "required_pages_per_second_upper_estimate": 0.004354973199155766
     },
     {
       "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": 28219.79231528734,
-      "required_pages_per_second_upper_estimate": 0.0017184372941932312
+      "required_token_positions_per_second": 28238.412698609038,
+      "required_pages_per_second_upper_estimate": 0.0017195711778439213
     },
     {
       "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": 28219.79231528734,
-      "required_pages_per_second_upper_estimate": 0.0021724377298156833
+      "required_token_positions_per_second": 28238.412698609038,
+      "required_pages_per_second_upper_estimate": 0.0021738711784683074
     },
     {
       "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": 28219.79231528734,
-      "required_pages_per_second_upper_estimate": 0.3503953749864949
+      "required_token_positions_per_second": 28238.412698609038,
+      "required_pages_per_second_upper_estimate": 0.35062657782893625
     }
   ],
   "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": 0.09057034419524257,
+    "producer_batch_seconds": 0.08369429818748131,
     "producer_p95_seconds": 0.21676743700299994,
-    "gpu_update_seconds": 0.9797378978236637,
-    "producer_batches_per_second": 11.041141654980345,
-    "gpu_batches_per_second": 1.0206811456628813,
-    "producer_headroom": 10.81742491462373,
+    "gpu_update_seconds": 0.9790918595563227,
+    "producer_batches_per_second": 11.948245240792,
+    "gpu_batches_per_second": 1.0213546259624218,
+    "producer_headroom": 11.698429651242021,
     "sustainable": true,
-    "buffer_seconds": 60.74374966506715,
+    "buffer_seconds": 60.70369529249201,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386194.5324039302,
-    "upload_bytes_per_second": 4789757.688183731,
+    "generation_bytes_per_second": 416675.2097426234,
+    "upload_bytes_per_second": 4810936.235122628,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.86835650528364,
+    "estimated_full_shards_per_hour": 89.40880030831362,
     "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": 0.7852815879996342,
-      "bytes": 838334,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 4.030324741001095,
       "bytes": 15386927,
@@ -562,6 +556,12 @@
       "bytes": 12098645,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.8189289659967471,
+      "bytes": 7675879,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -158,3 +158,4 @@
 {"step": 252900, "loss": 0.603326765820384, "lr": 0.0003, "grad_norm": 0.16471991318646134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37017867429, "tokens_per_second": 149367.99469737144, "data_wait_s": 1.319799957855139e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 253000, "loss": 0.6166946720331907, "lr": 0.0003, "grad_norm": 0.1520392208692349, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37032638328, "tokens_per_second": 151319.9703398773, "data_wait_s": 1.3380000382312573e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 253100, "loss": 0.5919720446690917, "lr": 0.0003, "grad_norm": 0.1526509503626916, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37047144594, "tokens_per_second": 148602.45958990487, "data_wait_s": 1.5008998161647469e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 253200, "loss": 0.5419284403324127, "lr": 0.0003, "grad_norm": 0.15212501898844985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37061814354, "tokens_per_second": 149998.83716176712, "data_wait_s": 1.1742999049602076e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bc422fced3930ce1b9dcb34db4043c0639f11c440d041b32f6091d8e526a703b
-size 341394
+oid sha256:4df4fc4eb4b1a858a1d30a19a9f7b034fb1c0d04c5c9befd325f36cf939ef489
+size 343512