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": 76176.29699155399,
-      "required_pages_per_second_upper_estimate": 0.005270052979236586
+      "required_token_positions_per_second": 76181.50536885101,
+      "required_pages_per_second_upper_estimate": 0.005270413306863095
     },
     {
       "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": 9522.037123944248,
-      "required_pages_per_second_upper_estimate": 0.003986678101049146
+      "required_token_positions_per_second": 9522.688171106376,
+      "required_pages_per_second_upper_estimate": 0.003986950680900519
     },
     {
       "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": 19044.074247888497,
-      "required_pages_per_second_upper_estimate": 0.0039663784398240205
+      "required_token_positions_per_second": 19045.376342212752,
+      "required_pages_per_second_upper_estimate": 0.0039666496317332265
     },
     {
       "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": 28566.111371832743,
-      "required_pages_per_second_upper_estimate": 0.001712494758814868
+      "required_token_positions_per_second": 28568.06451331913,
+      "required_pages_per_second_upper_estimate": 0.0017126118466647025
     },
     {
       "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": 28566.111371832743,
-      "required_pages_per_second_upper_estimate": 0.0022208895101150715
+      "required_token_positions_per_second": 28568.06451331913,
+      "required_pages_per_second_upper_estimate": 0.0022210413582746764
     },
     {
       "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": 28566.111371832743,
-      "required_pages_per_second_upper_estimate": 0.36641540477716733
+      "required_token_positions_per_second": 28568.06451331913,
+      "required_pages_per_second_upper_estimate": 0.36644045757903476
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.47207781937513005,
-    "producer_p95_seconds": 0.30505003899816074,
-    "gpu_update_seconds": 0.9678601206904893,
-    "producer_batches_per_second": 2.1182948212302346,
-    "gpu_batches_per_second": 1.033207153205756,
-    "producer_headroom": 2.050213081333933,
+    "producer_batch_seconds": 0.44204897689061795,
+    "producer_p95_seconds": 0.30792599900087225,
+    "gpu_update_seconds": 0.9677939500280052,
+    "producer_batches_per_second": 2.262192771113331,
+    "gpu_batches_per_second": 1.0332777963440078,
+    "producer_headroom": 2.18933647768057,
     "sustainable": true,
-    "buffer_seconds": 60.00732748281034,
+    "buffer_seconds": 60.00322490173632,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 302208.7562530897,
-    "upload_bytes_per_second": 5543887.316173678,
+    "generation_bytes_per_second": 280000.87900274777,
+    "upload_bytes_per_second": 5614891.931317678,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.84696403212088,
+    "estimated_full_shards_per_hour": 60.081670547121284,
     "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": 1.213385584000207,
-      "bytes": 6122336,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.3804073209976195,
       "bytes": 6371391,
@@ -562,6 +556,12 @@
       "bytes": 6071875,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.52896361799867,
+      "bytes": 13388189,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -319,3 +319,4 @@
 {"step": 676900, "loss": 0.5266819596290588, "lr": 0.0001, "grad_norm": 0.2042373696576169, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99080750240, "tokens_per_second": 152655.05764364393, "data_wait_s": 1.166099900729023e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 677000, "loss": 0.6073641180992126, "lr": 0.0001, "grad_norm": 0.20068715210851132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99095332637, "tokens_per_second": 146804.93471253722, "data_wait_s": 1.1740998161258176e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 677100, "loss": 0.5368040800094604, "lr": 0.0001, "grad_norm": 0.20854331524786404, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99110019550, "tokens_per_second": 152624.3591450172, "data_wait_s": 9.710001904750243e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 677200, "loss": 0.5721763968467712, "lr": 0.0001, "grad_norm": 0.18616434732814036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99124691637, "tokens_per_second": 152216.9666889091, "data_wait_s": 9.499999578110874e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:39401dfcc3d4cddf0c5d7816260f37b59a58c27c5ce24c7a2eb412f78d76da39
-size 684741
+oid sha256:1771a469d0a1f7e56a03ee48e162ef3983dce39793d5d77004ec66ee7c810f69
+size 686859