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": 76028.66542665874,
-      "required_pages_per_second_upper_estimate": 0.00658997202629194
+      "required_token_positions_per_second": 75538.99012791507,
+      "required_pages_per_second_upper_estimate": 0.006547528212467536
     },
     {
       "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": 9503.583178332343,
-      "required_pages_per_second_upper_estimate": 0.003791121421067633
+      "required_token_positions_per_second": 9442.373765989383,
+      "required_pages_per_second_upper_estimate": 0.003766704071321758
     },
     {
       "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": 19007.166356664686,
-      "required_pages_per_second_upper_estimate": 0.0043969734219115935
+      "required_token_positions_per_second": 18884.747531978766,
+      "required_pages_per_second_upper_estimate": 0.004368653981318235
     },
     {
       "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": 28510.74953499703,
-      "required_pages_per_second_upper_estimate": 0.001736155062338996
+      "required_token_positions_per_second": 28327.12129796815,
+      "required_pages_per_second_upper_estimate": 0.0017249730661268398
     },
     {
       "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": 28510.74953499703,
-      "required_pages_per_second_upper_estimate": 0.0021948364220100687
+      "required_token_positions_per_second": 28327.12129796815,
+      "required_pages_per_second_upper_estimate": 0.002180700212008092
     },
     {
       "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": 28510.74953499703,
-      "required_pages_per_second_upper_estimate": 0.35400808988411575
+      "required_token_positions_per_second": 28327.12129796815,
+      "required_pages_per_second_upper_estimate": 0.35172804174439265
     }
   ],
   "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": 0.11566469687494418,
-    "producer_p95_seconds": 0.28006331299911835,
-    "gpu_update_seconds": 0.9697395000458335,
-    "producer_batches_per_second": 8.645680376279312,
-    "gpu_batches_per_second": 1.0312047719544644,
-    "producer_headroom": 8.384057765649175,
+    "producer_batch_seconds": 0.06506710257053783,
+    "producer_p95_seconds": 0.2081564639993303,
+    "gpu_update_seconds": 0.9760257567006336,
+    "producer_batches_per_second": 15.368749498503053,
+    "gpu_batches_per_second": 1.0245631256498897,
+    "producer_headroom": 15.000295358818926,
     "sustainable": true,
-    "buffer_seconds": 60.12384900284168,
+    "buffer_seconds": 60.51359691543929,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 333985.0791744951,
-    "upload_bytes_per_second": 5169178.207380975,
+    "generation_bytes_per_second": 278024.65559413296,
+    "upload_bytes_per_second": 5136881.452576272,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.66542321611539,
+    "estimated_full_shards_per_hour": 59.65761900775902,
     "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": 1.169628391000515,
-      "bytes": 7496810,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.5263510319982743,
       "bytes": 5728692,
@@ -562,6 +556,12 @@
       "bytes": 7770781,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1696365809984854,
+      "bytes": 4882259,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -185,3 +185,4 @@
 {"step": 255600, "loss": 0.6161860916763544, "lr": 0.0003, "grad_norm": 0.14620557356839445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37412889586, "tokens_per_second": 149742.4806452215, "data_wait_s": 1.537800199002959e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 255700, "loss": 0.6096172984689474, "lr": 0.0003, "grad_norm": 0.17134667809357576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37427631977, "tokens_per_second": 150449.1907204958, "data_wait_s": 1.467999754822813e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 255800, "loss": 0.6042794194072485, "lr": 0.0003, "grad_norm": 0.14838269205064727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37442394541, "tokens_per_second": 140723.64876658912, "data_wait_s": 1.5637997421436012e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 255900, "loss": 0.6159732881933451, "lr": 0.0003, "grad_norm": 0.1648850653482918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37457210876, "tokens_per_second": 151658.94846547183, "data_wait_s": 1.2847998732468113e-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:f187a2ba7308fec7983c4e3baaa3a76de0f63795be98b44d64e79fc6b2ce39d0
-size 399363
+oid sha256:163b5b6ce5d211801dff7501cfd989bc2aa650102f8fb4d30fb541a28304e680
+size 401481