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": 73581.04013621887,
-      "required_pages_per_second_upper_estimate": 0.006377818069566196
+      "required_token_positions_per_second": 72711.50144584462,
+      "required_pages_per_second_upper_estimate": 0.006302448659710218
     },
     {
       "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": 9197.630017027359,
-      "required_pages_per_second_upper_estimate": 0.003669072130615669
+      "required_token_positions_per_second": 9088.937680730578,
+      "required_pages_per_second_upper_estimate": 0.0036257131325716364
     },
     {
       "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": 18395.260034054718,
-      "required_pages_per_second_upper_estimate": 0.004255419663359242
+      "required_token_positions_per_second": 18177.875361461156,
+      "required_pages_per_second_upper_estimate": 0.004205131545194291
     },
     {
       "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": 27592.89005108208,
-      "required_pages_per_second_upper_estimate": 0.0016802622354051156
+      "required_token_positions_per_second": 27266.813042191734,
+      "required_pages_per_second_upper_estimate": 0.0016604058563575422
     },
     {
       "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": 27592.89005108208,
-      "required_pages_per_second_upper_estimate": 0.0021241770581406257
+      "required_token_positions_per_second": 27266.813042191734,
+      "required_pages_per_second_upper_estimate": 0.002099074747357315
     },
     {
       "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": 27592.89005108208,
-      "required_pages_per_second_upper_estimate": 0.3426113469719766
+      "required_token_positions_per_second": 27266.813042191734,
+      "required_pages_per_second_upper_estimate": 0.3385625618311054
     }
   ],
   "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": 0.2908702613746641,
-    "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,
+    "producer_batch_seconds": 0.20694127872684476,
+    "producer_p95_seconds": 0.5213090020042728,
+    "gpu_update_seconds": 1.0139798867296457,
+    "producer_batches_per_second": 4.832288686685681,
+    "gpu_batches_per_second": 0.9862128559820506,
+    "producer_headroom": 4.899843535170494,
     "sustainable": true,
-    "buffer_seconds": 60.11983510712194,
+    "buffer_seconds": 60.83879320377874,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 372712.89785923244,
-    "upload_bytes_per_second": 4418342.884615282,
+    "generation_bytes_per_second": 370304.1963237445,
+    "upload_bytes_per_second": 4376941.838641347,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.97551156838159,
+    "estimated_full_shards_per_hour": 79.45866029056788,
     "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": 1.727812716999324,
-      "bytes": 13149585,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 0.9765169189995504,
       "bytes": 821604,
@@ -562,6 +556,12 @@
       "bytes": 5361652,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.3794091339950683,
+      "bytes": 7649365,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -414,3 +414,4 @@
 {"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}
+{"step": 278800, "loss": 0.6008367203176022, "lr": 0.0003, "grad_norm": 0.16020589139534827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40808645024, "tokens_per_second": 145526.61215152068, "data_wait_s": 1.867199898697436e-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 sha256:56e43c67957e134cca3d2834afe19e283d4704e85ac4ded5ddb5d6d3d833d8c3
-size 887517
+oid sha256:72dceeee414a33ace92d6d30e1840c7e0d1eb20fcc8b64c63512c2d289446b46
+size 889635