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": 75354.42164128022,
-      "required_pages_per_second_upper_estimate": 0.00653153028383058
+      "required_token_positions_per_second": 74965.3973539503,
+      "required_pages_per_second_upper_estimate": 0.006497810644577866
     },
     {
       "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": 9419.302705160027,
-      "required_pages_per_second_upper_estimate": 0.0037575006802138293
+      "required_token_positions_per_second": 9370.674669243788,
+      "required_pages_per_second_upper_estimate": 0.003738102229632914
     },
     {
       "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": 18838.605410320055,
-      "required_pages_per_second_upper_estimate": 0.004357979813546098
+      "required_token_positions_per_second": 18741.349338487576,
+      "required_pages_per_second_upper_estimate": 0.004335481332976058
     },
     {
       "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": 28257.908115480084,
-      "required_pages_per_second_upper_estimate": 0.0017207583464468206
+      "required_token_positions_per_second": 28112.02400773137,
+      "required_pages_per_second_upper_estimate": 0.0017118747696796843
     },
     {
       "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": 28257.908115480084,
-      "required_pages_per_second_upper_estimate": 0.0021753719896258085
+      "required_token_positions_per_second": 28112.02400773137,
+      "required_pages_per_second_upper_estimate": 0.002164141427178256
     },
     {
       "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": 28257.908115480084,
-      "required_pages_per_second_upper_estimate": 0.3508686456595116
+      "required_token_positions_per_second": 28112.02400773137,
+      "required_pages_per_second_upper_estimate": 0.3490572532839734
     }
   ],
   "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": 0.13859298968731082,
-    "producer_p95_seconds": 0.45681583599798614,
-    "gpu_update_seconds": 0.9784163741708125,
-    "producer_batches_per_second": 7.215372164610698,
-    "gpu_batches_per_second": 1.0220597553342043,
-    "producer_headroom": 7.059638271591406,
+    "producer_batch_seconds": 0.5699123814143832,
+    "producer_p95_seconds": 0.45538208500511246,
+    "gpu_update_seconds": 0.9834937531497643,
+    "producer_batches_per_second": 1.7546556849988844,
+    "gpu_batches_per_second": 1.0167832757425985,
+    "producer_headroom": 1.7256929051251235,
     "sustainable": true,
-    "buffer_seconds": 60.661815198590375,
+    "buffer_seconds": 60.97661269528538,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 292002.11942651926,
-    "upload_bytes_per_second": 6169005.99111494,
+    "generation_bytes_per_second": 261433.70408706358,
+    "upload_bytes_per_second": 6253193.739659444,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.65685736748394,
+    "estimated_full_shards_per_hour": 56.097587031926444,
     "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": 0.8141304599994328,
-      "bytes": 2723440,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 0.9625181489973329,
       "bytes": 5689945,
@@ -562,6 +556,12 @@
       "bytes": 5901383,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.330521811003564,
+      "bytes": 11596177,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -415,3 +415,4 @@
 {"step": 81500, "loss": 0.6229772474616766, "lr": 0.0003, "grad_norm": 0.13445697579748267, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11927072858, "tokens_per_second": 148503.2866475685, "data_wait_s": 1.4524994185194373e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 81600, "loss": 0.6269095726311207, "lr": 0.0003, "grad_norm": 0.13646336345421883, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11941723632, "tokens_per_second": 148375.01129323352, "data_wait_s": 8.81299638422206e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 81700, "loss": 0.7072742599993944, "lr": 0.0003, "grad_norm": 0.16792434377732804, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11956176838, "tokens_per_second": 145960.47523637753, "data_wait_s": 1.1839998478535563e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 81800, "loss": 0.6340364888310432, "lr": 0.0003, "grad_norm": 0.14381530252142383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11970645618, "tokens_per_second": 129867.3593586112, "data_wait_s": 1.4462995750363916e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:609aab0b02238808b37da11a65284fa62846bbb112ed3f154db60bcbfe1c1806
-size 889635
+oid sha256:0931769004599d9e2ef5449a7d92502afc604802b2024e3a6eab9cf6f2451d98
+size 891753