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": 76915.99263091253,
-      "required_pages_per_second_upper_estimate": 0.005321226839897757
+      "required_token_positions_per_second": 77226.89891718194,
+      "required_pages_per_second_upper_estimate": 0.005342736058183849
     },
     {
       "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": 9614.499078864066,
-      "required_pages_per_second_upper_estimate": 0.004025389990748894
+      "required_token_positions_per_second": 9653.362364647743,
+      "required_pages_per_second_upper_estimate": 0.004041661236948827
     },
     {
       "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": 19228.998157728132,
-      "required_pages_per_second_upper_estimate": 0.004004893213472166
+      "required_token_positions_per_second": 19306.724729295485,
+      "required_pages_per_second_upper_estimate": 0.004021081608542111
     },
     {
       "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": 28843.497236592197,
-      "required_pages_per_second_upper_estimate": 0.0017291236178635057
+      "required_token_positions_per_second": 28960.087093943228,
+      "required_pages_per_second_upper_estimate": 0.0017361130017892958
     },
     {
       "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": 28843.497236592197,
-      "required_pages_per_second_upper_estimate": 0.002242455041008643
+      "required_token_positions_per_second": 28960.087093943228,
+      "required_pages_per_second_upper_estimate": 0.0022515193895931004
     },
     {
       "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": 28843.497236592197,
-      "required_pages_per_second_upper_estimate": 0.36997341281656465
+      "required_token_positions_per_second": 28960.087093943228,
+      "required_pages_per_second_upper_estimate": 0.3714689023222281
     }
   ],
   "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": 0.08503389130478922,
-    "producer_p95_seconds": 0.2274037090028287,
-    "gpu_update_seconds": 0.9585522786371573,
-    "producer_batches_per_second": 11.760016913911109,
-    "gpu_batches_per_second": 1.0432399174114655,
-    "producer_headroom": 11.272591009641003,
+    "producer_batch_seconds": 0.08077422254703492,
+    "producer_p95_seconds": 0.22166756899969187,
+    "gpu_update_seconds": 0.9546932614640637,
+    "producer_batches_per_second": 12.380187248694332,
+    "gpu_batches_per_second": 1.0474568538029234,
+    "producer_headroom": 11.819281341991806,
     "sustainable": true,
-    "buffer_seconds": 60.38879355414091,
+    "buffer_seconds": 60.145675472236015,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 286260.7919257086,
-    "upload_bytes_per_second": 5484645.514540677,
+    "generation_bytes_per_second": 274060.72807406256,
+    "upload_bytes_per_second": 5528379.080066351,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.42490213707393,
+    "estimated_full_shards_per_hour": 58.80705243746193,
     "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": 1.0711626960001013,
-      "bytes": 2631436,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.137612402000741,
       "bytes": 5722821,
@@ -562,6 +556,12 @@
       "bytes": 13264605,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.0446096289997513,
+      "bytes": 5930135,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -200,3 +200,4 @@
 {"step": 545000, "loss": 0.5358173847198486, "lr": 0.0003, "grad_norm": 0.16073940876093196, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79770616642, "tokens_per_second": 152186.375552202, "data_wait_s": 8.464998245472088e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 545100, "loss": 0.5718283653259277, "lr": 0.0003, "grad_norm": 0.16655540749504766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79785257676, "tokens_per_second": 141356.07232460906, "data_wait_s": 1.2747997971018776e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 545200, "loss": 0.5898277163505554, "lr": 0.0003, "grad_norm": 0.15976357095997862, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79799850845, "tokens_per_second": 152420.8837662645, "data_wait_s": 1.3625998690258712e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 545300, "loss": 0.5992398262023926, "lr": 0.0003, "grad_norm": 0.15674889889811952, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79814546569, "tokens_per_second": 153491.28211642875, "data_wait_s": 9.816001693252474e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e11d0f828736704194b989792f4743053b5bdd62c1fb7b1e0ce863133a15d675
-size 431133
+oid sha256:ae3e2b258cbe0f694692ba40a312087ff428c133b242a41ea5a7d9773b46bcd0
+size 433251