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": 76203.51120805221,
-      "required_pages_per_second_upper_estimate": 0.005271935721879615
+      "required_token_positions_per_second": 76587.58244272173,
+      "required_pages_per_second_upper_estimate": 0.005298506661062096
     },
     {
       "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": 9525.438901006526,
-      "required_pages_per_second_upper_estimate": 0.003988102354067939
+      "required_token_positions_per_second": 9573.447805340216,
+      "required_pages_per_second_upper_estimate": 0.00400820267977253
     },
     {
       "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": 19050.877802013052,
-      "required_pages_per_second_upper_estimate": 0.003967795440726378
+      "required_token_positions_per_second": 19146.895610680433,
+      "required_pages_per_second_upper_estimate": 0.003987793418111898
     },
     {
       "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": 28576.316703019576,
-      "required_pages_per_second_upper_estimate": 0.0017131065528368803
+      "required_token_positions_per_second": 28720.343416020653,
+      "required_pages_per_second_upper_estimate": 0.0017217407343652375
     },
     {
       "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": 28576.316703019576,
-      "required_pages_per_second_upper_estimate": 0.0022216829297263397
+      "required_token_positions_per_second": 28720.343416020653,
+      "required_pages_per_second_upper_estimate": 0.002232880373155614
     },
     {
       "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": 28576.316703019576,
-      "required_pages_per_second_upper_estimate": 0.3665463078080011
+      "required_token_positions_per_second": 28720.343416020653,
+      "required_pages_per_second_upper_estimate": 0.368393727838543
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.0797802882577372,
-    "producer_p95_seconds": 0.22324974600087444,
-    "gpu_update_seconds": 0.9675144731678633,
-    "producer_batches_per_second": 12.534424503072897,
-    "gpu_batches_per_second": 1.0335762696404651,
-    "producer_headroom": 12.12723711955293,
+    "producer_batch_seconds": 0.06605057153119276,
+    "producer_p95_seconds": 0.22654985900044267,
+    "gpu_update_seconds": 0.9626625837829473,
+    "producer_batches_per_second": 15.139914414332422,
+    "gpu_batches_per_second": 1.0387855691558394,
+    "producer_headroom": 14.574629128353937,
     "sustainable": true,
-    "buffer_seconds": 60.95341180957539,
+    "buffer_seconds": 60.64774277832568,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317982.8262899309,
-    "upload_bytes_per_second": 5286994.596617914,
+    "generation_bytes_per_second": 317908.44062442693,
+    "upload_bytes_per_second": 5284767.532284105,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.23171226046986,
+    "estimated_full_shards_per_hour": 68.21575082826239,
     "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": 1.2697396670000671,
-      "bytes": 6182466,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.2363136280000617,
       "bytes": 6022734,
@@ -562,6 +556,12 @@
       "bytes": 6660673,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.2537715590005973,
+      "bytes": 5914249,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -150,3 +150,4 @@
 {"step": 900000, "loss": 0.5382843017578125, "lr": 3e-05, "grad_norm": 0.327387548139761, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131734420906, "tokens_per_second": 134785.8818095809, "data_wait_s": 8.85099871084094e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 900100, "loss": 0.5342219471931458, "lr": 3e-05, "grad_norm": 0.3243542243782647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131749066860, "tokens_per_second": 134505.8787989976, "data_wait_s": 9.325000064563937e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 900200, "loss": 0.5399182438850403, "lr": 3e-05, "grad_norm": 0.3283343396842356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131763803760, "tokens_per_second": 153995.44107381676, "data_wait_s": 1.1669999366858974e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 900300, "loss": 0.5195290446281433, "lr": 3e-05, "grad_norm": 0.30953746353027317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131778484935, "tokens_per_second": 153178.59357082617, "data_wait_s": 2.0464000044739805e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:60e53d39d0718c3f3eed0db1d49374ad4809946aa5e508e209eae051c812b320
-size 324450
+oid sha256:ff55d17f72960738ef29fb1efb136df8e2c9905c1effbd78ef8efa5c954d230c
+size 326568