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": 75597.86980031131,
-      "required_pages_per_second_upper_estimate": 0.005230036044010423
+      "required_token_positions_per_second": 73791.83096575811,
+      "required_pages_per_second_upper_estimate": 0.005105090086848587
     },
     {
       "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": 9449.733725038914,
-      "required_pages_per_second_upper_estimate": 0.003956406177794145
+      "required_token_positions_per_second": 9223.978870719764,
+      "required_pages_per_second_upper_estimate": 0.0038618873345881554
     },
     {
       "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": 18899.467450077827,
-      "required_pages_per_second_upper_estimate": 0.0039362606573777655
+      "required_token_positions_per_second": 18447.957741439528,
+      "required_pages_per_second_upper_estimate": 0.0038422230921801434
     },
     {
       "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": 28349.201175116745,
-      "required_pages_per_second_upper_estimate": 0.001699491323724437
+      "required_token_positions_per_second": 27671.93661215929,
+      "required_pages_per_second_upper_estimate": 0.0016588903472982476
     },
     {
       "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": 28349.201175116745,
-      "required_pages_per_second_upper_estimate": 0.0022040256964075268
+      "required_token_positions_per_second": 27671.93661215929,
+      "required_pages_per_second_upper_estimate": 0.00215137135560957
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28349.201175116745,
-      "required_pages_per_second_upper_estimate": 0.3636331136737182
+      "required_token_positions_per_second": 27671.93661215929,
+      "required_pages_per_second_upper_estimate": 0.35494589105013136
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12715467205484288,
-    "producer_p95_seconds": 0.31598223599939956,
-    "gpu_update_seconds": 0.9752655755347274,
-    "producer_batches_per_second": 7.864437726430466,
-    "gpu_batches_per_second": 1.0253617323175905,
-    "producer_headroom": 7.669915385524231,
+    "producer_batch_seconds": 0.2562649832969157,
+    "producer_p95_seconds": 0.3117504590009048,
+    "gpu_update_seconds": 0.9991349860150817,
+    "producer_batches_per_second": 3.9022108566482228,
+    "gpu_batches_per_second": 1.0008657628819189,
+    "producer_headroom": 3.898835389685122,
     "sustainable": true,
-    "buffer_seconds": 60.466465683153096,
+    "buffer_seconds": 60.94723414691998,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 484630.29120586516,
-    "upload_bytes_per_second": 4285666.65230903,
+    "generation_bytes_per_second": 460345.019986264,
+    "upload_bytes_per_second": 4185822.474152754,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.99037887699095,
+    "estimated_full_shards_per_hour": 98.7793250054449,
     "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": 1.6203178860014305,
-      "bytes": 6721657,
-      "pages": 29,
-      "shards": 3
-    },
-    {
-      "seconds": 1.8514023890020326,
-      "bytes": 12113021,
-      "pages": 32,
-      "shards": 4
-    },
     {
       "seconds": 1.0979482909970102,
       "bytes": 1151793,
@@ -562,6 +550,18 @@
       "bytes": 5846921,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 2.7826384430009057,
+      "bytes": 7726579,
+      "pages": 35,
+      "shards": 3
+    },
+    {
+      "seconds": 1.4537843559992325,
+      "bytes": 4298944,
+      "pages": 20,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -323,3 +323,4 @@
 {"step": 874800, "loss": 0.5186391472816467, "lr": 3e-05, "grad_norm": 0.3152294704003549, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128046110866, "tokens_per_second": 152940.05195945117, "data_wait_s": 1.3041997590335086e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 874900, "loss": 0.537115216255188, "lr": 3e-05, "grad_norm": 0.30592999147303934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128060867543, "tokens_per_second": 154272.42569300413, "data_wait_s": 8.540002454537898e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 875000, "loss": 0.5282732248306274, "lr": 3e-05, "grad_norm": 0.30791818889206707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128075402216, "tokens_per_second": 150694.60485398571, "data_wait_s": 1.2407999747665599e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 875100, "loss": 0.4834609925746918, "lr": 3e-05, "grad_norm": 0.3024347114563917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128090022645, "tokens_per_second": 139551.35050221166, "data_wait_s": 1.2865999451605603e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:811d4161236a98fda25bdc19a474d1d5725c4209ea0c9ad6dfed6e21decd1529
-size 693996
+oid sha256:7a92cdca6dcde36b724284863843d2f7108e8d6162c812972e78474534c19c22
+size 696114