Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76608.2440593413,
-      "required_pages_per_second_upper_estimate": 0.005299936079641394
+      "required_token_positions_per_second": 77363.2261112717,
+      "required_pages_per_second_upper_estimate": 0.00535216748979365
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9576.030507417663,
-      "required_pages_per_second_upper_estimate": 0.004009284003199404
+      "required_token_positions_per_second": 9670.403263908962,
+      "required_pages_per_second_upper_estimate": 0.004048795905615057
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19152.061014835326,
-      "required_pages_per_second_upper_estimate": 0.003988869235576494
+      "required_token_positions_per_second": 19340.806527817924,
+      "required_pages_per_second_upper_estimate": 0.004028179948376866
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28728.09152225299,
-      "required_pages_per_second_upper_estimate": 0.001722205221503193
+      "required_token_positions_per_second": 29011.209791726884,
+      "required_pages_per_second_upper_estimate": 0.0017391777294615861
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28728.09152225299,
-      "required_pages_per_second_upper_estimate": 0.002233482754334861
+      "required_token_positions_per_second": 29011.209791726884,
+      "required_pages_per_second_upper_estimate": 0.0022554939544807964
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28728.09152225299,
-      "required_pages_per_second_upper_estimate": 0.36849311222602316
+      "required_token_positions_per_second": 29011.209791726884,
+      "required_pages_per_second_upper_estimate": 0.3721246493981206
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.6385047882343144,
-    "producer_p95_seconds": 0.21890624099978595,
-    "gpu_update_seconds": 0.962402948994494,
-    "producer_batches_per_second": 1.5661589676803274,
-    "gpu_batches_per_second": 1.0390658102666737,
-    "producer_headroom": 1.5072760090897195,
+    "producer_batch_seconds": 0.060190696156354306,
+    "producer_p95_seconds": 0.2101595099993574,
+    "gpu_update_seconds": 0.9530109291713981,
+    "producer_batches_per_second": 16.61386333532928,
+    "gpu_batches_per_second": 1.0493059097123438,
+    "producer_headroom": 15.833193334328783,
     "sustainable": true,
-    "buffer_seconds": 60.63138578665312,
+    "buffer_seconds": 60.03968853779808,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 310794.11682017095,
-    "upload_bytes_per_second": 5262266.363159667,
+    "generation_bytes_per_second": 310341.4372502134,
+    "upload_bytes_per_second": 5363215.1755283605,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.68918255285116,
+    "estimated_full_shards_per_hour": 66.59204805497934,
     "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": 1.4154219219999504,
-      "bytes": 7407761,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.1865251550007088,
       "bytes": 5346992,
@@ -562,6 +556,12 @@
       "bytes": 2670941,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.4827766799971869,
+      "bytes": 15946512,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -177,3 +177,4 @@
 {"step": 574800, "loss": 0.7473382353782654, "lr": 0.0003, "grad_norm": 0.3193019051904646, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84133290730, "tokens_per_second": 152766.53100402327, "data_wait_s": 1.2799999240087345e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 574900, "loss": 0.5708333849906921, "lr": 0.0003, "grad_norm": 0.1551997033269296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84148032987, "tokens_per_second": 154211.103268809, "data_wait_s": 8.013001206563786e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 575000, "loss": 0.6247869729995728, "lr": 0.0003, "grad_norm": 0.1688802240902574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84162817785, "tokens_per_second": 153800.6140100285, "data_wait_s": 8.392999006900936e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 575100, "loss": 0.6393929123878479, "lr": 0.0003, "grad_norm": 0.1982794203423314, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84177504106, "tokens_per_second": 141397.47281022574, "data_wait_s": 1.3353001122595742e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:80f4dd9a5a1987ed1079bd7ed19eac9380f909817fd742f26bcd50a7f207ab9b
-size 382419
+oid sha256:06a39e80831d4229da5790bc8d2dc5a5852c45123e3d9f2b99c157be0564be7a
+size 384537