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": 77084.99674134483,
-      "required_pages_per_second_upper_estimate": 0.005332918936401027
+      "required_token_positions_per_second": 76228.668914566,
+      "required_pages_per_second_upper_estimate": 0.005273676190390161
     },
     {
       "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": 9635.624592668104,
-      "required_pages_per_second_upper_estimate": 0.0040342348022277515
+      "required_token_positions_per_second": 9528.58361432075,
+      "required_pages_per_second_upper_estimate": 0.0039894189798635234
     },
     {
       "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": 19271.249185336208,
-      "required_pages_per_second_upper_estimate": 0.004013692988288401
+      "required_token_positions_per_second": 19057.1672286415,
+      "required_pages_per_second_upper_estimate": 0.00396910536242975
     },
     {
       "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": 28906.873778004312,
-      "required_pages_per_second_upper_estimate": 0.0017329229447507865
+      "required_token_positions_per_second": 28585.750842962254,
+      "required_pages_per_second_upper_estimate": 0.0017136721151213461
     },
     {
       "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": 28906.873778004312,
-      "required_pages_per_second_upper_estimate": 0.0022473822883394903
+      "required_token_positions_per_second": 28585.750842962254,
+      "required_pages_per_second_upper_estimate": 0.0022224163926104867
     },
     {
       "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": 28906.873778004312,
-      "required_pages_per_second_upper_estimate": 0.3707863390413708
+      "required_token_positions_per_second": 28585.750842962254,
+      "required_pages_per_second_upper_estimate": 0.36666731882559556
     }
   ],
   "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": 0.14365813682030648,
-    "producer_p95_seconds": 0.32586697800434195,
-    "gpu_update_seconds": 0.9564507117693851,
-    "producer_batches_per_second": 6.960970134610901,
-    "gpu_batches_per_second": 1.0455321823641606,
-    "producer_headroom": 6.657824839854029,
+    "producer_batch_seconds": 0.40795902450020094,
+    "producer_p95_seconds": 0.32263869399321266,
+    "gpu_update_seconds": 0.9671951648877845,
+    "producer_batches_per_second": 2.451226569200475,
+    "gpu_batches_per_second": 1.0339174928733454,
+    "producer_headroom": 2.3708144857751714,
     "sustainable": true,
-    "buffer_seconds": 60.25639484147126,
+    "buffer_seconds": 60.93329538793042,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 338873.2052455837,
-    "upload_bytes_per_second": 4854826.454614653,
+    "generation_bytes_per_second": 329716.1421738058,
+    "upload_bytes_per_second": 4878668.169995524,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.7143012812198,
+    "estimated_full_shards_per_hour": 70.74940871153478,
     "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": 1.2067919649998657,
-      "bytes": 5662165,
-      "pages": 21,
-      "shards": 2
-    },
     {
       "seconds": 1.381297531999735,
       "bytes": 11693702,
@@ -562,6 +556,12 @@
       "bytes": 887251,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.1445283560024109,
+      "bytes": 7481150,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -366,3 +366,4 @@
 {"step": 796600, "loss": 0.520895779132843, "lr": 3e-05, "grad_norm": 0.25431231634459683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116599788305, "tokens_per_second": 153603.92232805223, "data_wait_s": 1.2486001651268452e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 796700, "loss": 0.5909894108772278, "lr": 3e-05, "grad_norm": 0.28662822347652184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116614287455, "tokens_per_second": 150561.0224878996, "data_wait_s": 1.1718999303411692e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 796800, "loss": 0.5350161790847778, "lr": 3e-05, "grad_norm": 0.26982433464996836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116628817201, "tokens_per_second": 150991.49962865576, "data_wait_s": 1.1568998161237687e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 796900, "loss": 0.48557382822036743, "lr": 3e-05, "grad_norm": 0.26081033136248116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116643322906, "tokens_per_second": 151187.62686386445, "data_wait_s": 1.1613999959081411e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d09967df9b888b148f7b7d14fd679df6c64e14ed2224c7a60d07eb0fced5ecdc
-size 785070
+oid sha256:3ebad3bf84aa1fb4f47675c48db9ca6af564a5230f2d2faf03384dabd87c5ac3
+size 787188