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": 76596.21227283601,
-      "required_pages_per_second_upper_estimate": 0.005299103692733367
+      "required_token_positions_per_second": 77128.24863329732,
+      "required_pages_per_second_upper_estimate": 0.005335911202644516
     },
     {
       "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": 9574.526534104501,
-      "required_pages_per_second_upper_estimate": 0.004008654320979718
+      "required_token_positions_per_second": 9641.031079162165,
+      "required_pages_per_second_upper_estimate": 0.004036498385222538
     },
     {
       "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": 19149.053068209003,
-      "required_pages_per_second_upper_estimate": 0.003988242759619118
+      "required_token_positions_per_second": 19282.06215832433,
+      "required_pages_per_second_upper_estimate": 0.0040159450454045524
     },
     {
       "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": 28723.579602313504,
-      "required_pages_per_second_upper_estimate": 0.0017219347387921237
+      "required_token_positions_per_second": 28923.093237486497,
+      "required_pages_per_second_upper_estimate": 0.0017338952765810628
     },
     {
       "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": 28723.579602313504,
-      "required_pages_per_second_upper_estimate": 0.002233131972405404
+      "required_token_positions_per_second": 28923.093237486497,
+      "required_pages_per_second_upper_estimate": 0.0022486432799723676
     },
     {
       "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": 28723.579602313504,
-      "required_pages_per_second_upper_estimate": 0.3684352381615616
+      "required_token_positions_per_second": 28923.093237486497,
+      "required_pages_per_second_upper_estimate": 0.3709943848525096
     }
   ],
   "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": 0.09211662122658026,
-    "producer_p95_seconds": 0.23082868500205223,
-    "gpu_update_seconds": 0.9625541239216969,
-    "producer_batches_per_second": 10.855804160904784,
-    "gpu_batches_per_second": 1.0389026187179364,
-    "producer_headroom": 10.449299063565217,
+    "producer_batch_seconds": 0.16861545439826386,
+    "producer_p95_seconds": 0.23194225399856805,
+    "gpu_update_seconds": 0.955914354422027,
+    "producer_batches_per_second": 5.930654479855889,
+    "gpu_batches_per_second": 1.0461188236938115,
+    "producer_headroom": 5.669197748411545,
     "sustainable": true,
-    "buffer_seconds": 60.640909807066905,
+    "buffer_seconds": 60.2226043285877,
     "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": 342195.1805033183,
-    "upload_bytes_per_second": 4848505.4618605925,
+    "generation_bytes_per_second": 342824.15874215215,
+    "upload_bytes_per_second": 4791392.151073171,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.42711983990347,
+    "estimated_full_shards_per_hour": 73.56208392809319,
     "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": 1.2708631089990376,
-      "bytes": 9724430,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.795558247000372,
       "bytes": 12019830,
@@ -562,6 +556,12 @@
       "bytes": 6120614,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.370815718000813,
+      "bytes": 5076360,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -217,3 +217,4 @@
 {"step": 946700, "loss": 0.5465481877326965, "lr": 3e-05, "grad_norm": 0.34224239445298343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138570649993, "tokens_per_second": 151297.40515588396, "data_wait_s": 1.3739001587964594e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 946800, "loss": 0.49404534697532654, "lr": 3e-05, "grad_norm": 0.35579109683225085, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138585287594, "tokens_per_second": 152540.61911892937, "data_wait_s": 1.336399873252958e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 946900, "loss": 0.5574998259544373, "lr": 3e-05, "grad_norm": 0.33907311921784394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138600019929, "tokens_per_second": 153279.65545405858, "data_wait_s": 1.207899913424626e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 947000, "loss": 0.5088230967521667, "lr": 3e-05, "grad_norm": 0.32660654293101216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138614775562, "tokens_per_second": 154054.09789012928, "data_wait_s": 1.0375002602813765e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a7bafd038c1764ebfa4e9ff006abfbec44d76beed6a5f0f62e5aeac08db39af8
-size 467139
+oid sha256:abbdd073b4f1f46e148466023cfd605dab3e6eb93b00d1501068ea42ddded74a
+size 469257