Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 72819.7990437826,
-      "required_pages_per_second_upper_estimate": 0.006311835621021729
+      "required_token_positions_per_second": 73581.04013621887,
+      "required_pages_per_second_upper_estimate": 0.006377818069566196
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9102.474880472824,
-      "required_pages_per_second_upper_estimate": 0.0036311133239479913
+      "required_token_positions_per_second": 9197.630017027359,
+      "required_pages_per_second_upper_estimate": 0.003669072130615669
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18204.94976094565,
-      "required_pages_per_second_upper_estimate": 0.004211394731021871
+      "required_token_positions_per_second": 18395.260034054718,
+      "required_pages_per_second_upper_estimate": 0.004255419663359242
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27307.42464141847,
-      "required_pages_per_second_upper_estimate": 0.0016628788896778564
+      "required_token_positions_per_second": 27592.89005108208,
+      "required_pages_per_second_upper_estimate": 0.0016802622354051156
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27307.42464141847,
-      "required_pages_per_second_upper_estimate": 0.002102201140685891
+      "required_token_positions_per_second": 27592.89005108208,
+      "required_pages_per_second_upper_estimate": 0.0021241770581406257
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27307.42464141847,
-      "required_pages_per_second_upper_estimate": 0.3390668219752222
+      "required_token_positions_per_second": 27592.89005108208,
+      "required_pages_per_second_upper_estimate": 0.3426113469719766
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.1522657163206418,
-    "producer_p95_seconds": 0.4837486499964143,
-    "gpu_update_seconds": 1.012471895942357,
-    "producer_batches_per_second": 6.567466558881815,
-    "gpu_batches_per_second": 0.9876817361624157,
-    "producer_headroom": 6.649375318409099,
+    "producer_batch_seconds": 0.2908702613746641,
+    "producer_p95_seconds": 0.515428664002684,
+    "gpu_update_seconds": 1.0019972517853657,
+    "producer_batches_per_second": 3.4379588868039015,
+    "gpu_batches_per_second": 0.9980067292781423,
+    "producer_headroom": 3.444825356328584,
     "sustainable": true,
-    "buffer_seconds": 60.74831375654143,
+    "buffer_seconds": 60.11983510712194,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 381632.9798098704,
-    "upload_bytes_per_second": 4452507.604640421,
+    "generation_bytes_per_second": 372712.89785923244,
+    "upload_bytes_per_second": 4418342.884615282,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.88955350610813,
+    "estimated_full_shards_per_hour": 79.97551156838159,
     "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": 1.5038225289972615,
-      "bytes": 10256540,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.727812716999324,
       "bytes": 13149585,
@@ -562,6 +556,12 @@
       "bytes": 11687780,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.1412328320002416,
+      "bytes": 5361652,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -413,3 +413,4 @@
 {"step": 278400, "loss": 0.6388246286660433, "lr": 0.0003, "grad_norm": 0.16316806816724105, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40750257836, "tokens_per_second": 147282.42974894305, "data_wait_s": 1.0408999514766037e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 278500, "loss": 0.6065518260002136, "lr": 0.0003, "grad_norm": 0.1700394642829699, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40765085288, "tokens_per_second": 146634.99090987313, "data_wait_s": 1.2661002983804792e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 278600, "loss": 0.5725674219429493, "lr": 0.0003, "grad_norm": 0.13514279536204843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40779598508, "tokens_per_second": 145690.4113849115, "data_wait_s": 2.2104002709966153e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 278700, "loss": 0.5751864071935415, "lr": 0.0003, "grad_norm": 0.15900936643717134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40794039350, "tokens_per_second": 143447.7581870635, "data_wait_s": 1.870400592451915e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:03134a898ba73862744eed448f2a925dac42a3a289cbc4e045f01ac3603056f9
-size 885399
+oid sha256:56e43c67957e134cca3d2834afe19e283d4704e85ac4ded5ddb5d6d3d833d8c3
+size 887517