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": 77166.53355513202,
-      "required_pages_per_second_upper_estimate": 0.005338559842370298
+      "required_token_positions_per_second": 77195.70429665419,
+      "required_pages_per_second_upper_estimate": 0.005340577941954248
     },
     {
       "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": 9645.816694391502,
-      "required_pages_per_second_upper_estimate": 0.004038502022384052
+      "required_token_positions_per_second": 9649.463037081774,
+      "required_pages_per_second_upper_estimate": 0.0040400286699241745
     },
     {
       "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": 19291.633388783004,
-      "required_pages_per_second_upper_estimate": 0.004017938480298773
+      "required_token_positions_per_second": 19298.926074163548,
+      "required_pages_per_second_upper_estimate": 0.004019457354342494
     },
     {
       "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": 28937.450083174506,
-      "required_pages_per_second_upper_estimate": 0.0017347559475584249
+      "required_token_positions_per_second": 28948.389111245324,
+      "required_pages_per_second_upper_estimate": 0.0017354117255883417
     },
     {
       "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": 28937.450083174506,
-      "required_pages_per_second_upper_estimate": 0.0022497594615755196
+      "required_token_positions_per_second": 28948.389111245324,
+      "required_pages_per_second_upper_estimate": 0.0022506099228923266
     },
     {
       "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": 28937.450083174506,
-      "required_pages_per_second_upper_estimate": 0.3711785390538154
+      "required_token_positions_per_second": 28948.389111245324,
+      "required_pages_per_second_upper_estimate": 0.3713188531604947
     }
   ],
   "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": 0.2232352203593706,
-    "producer_p95_seconds": 0.2096004830000311,
-    "gpu_update_seconds": 0.955440093046614,
-    "producer_batches_per_second": 4.479579872701855,
-    "gpu_batches_per_second": 1.0466380961796335,
-    "producer_headroom": 4.2799702103840005,
+    "producer_batch_seconds": 0.06602440486719274,
+    "producer_p95_seconds": 0.19781329899888078,
+    "gpu_update_seconds": 0.955079050988016,
+    "producer_batches_per_second": 15.145914635830302,
+    "gpu_batches_per_second": 1.0470337496833522,
+    "producer_headroom": 14.465545776734308,
     "sustainable": true,
-    "buffer_seconds": 60.19272586193669,
+    "buffer_seconds": 60.16998021224501,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 383374.3955159706,
-    "upload_bytes_per_second": 3558364.8865642147,
+    "generation_bytes_per_second": 382814.02342733793,
+    "upload_bytes_per_second": 3631256.191271894,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.26322077855433,
+    "estimated_full_shards_per_hour": 82.14297797312835,
     "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": 1.0323733719997108,
-      "bytes": 1179407,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.2691897260001497,
       "bytes": 7296014,
@@ -562,6 +556,12 @@
       "bytes": 2716683,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3286046470002475,
+      "bytes": 11093643,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -142,3 +142,4 @@
 {"step": 939200, "loss": 0.5824522972106934, "lr": 3e-05, "grad_norm": 0.33713556494610447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137472164005, "tokens_per_second": 153448.18613484022, "data_wait_s": 6.291000318014994e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 939300, "loss": 0.5009631514549255, "lr": 3e-05, "grad_norm": 0.3215419116093088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137486866030, "tokens_per_second": 153759.34897092092, "data_wait_s": 1.393100137647707e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 939400, "loss": 0.515972375869751, "lr": 3e-05, "grad_norm": 0.3337855590228506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137501528991, "tokens_per_second": 153354.75197790645, "data_wait_s": 1.4064000424696133e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 939500, "loss": 0.5764984488487244, "lr": 3e-05, "grad_norm": 0.382731103550715, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137516223964, "tokens_per_second": 153219.29961798308, "data_wait_s": 1.6154999684658833e-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:cfe720454603e8b117fb7930205316767240a6cfa959f1b5e331cee6f59b60be
-size 306723
+oid sha256:e8b3d6a4cfb49c0f8a1b9361fe21e46b21834bf8c437ea8893bd0aca42f9a2b9
+size 308841