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": 75833.86787994622,
-      "required_pages_per_second_upper_estimate": 0.006573087468652767
+      "required_token_positions_per_second": 75912.4606356097,
+      "required_pages_per_second_upper_estimate": 0.00657989968952216
     },
     {
       "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": 9479.233484993278,
-      "required_pages_per_second_upper_estimate": 0.0037814079643343217
+      "required_token_positions_per_second": 9489.057579451213,
+      "required_pages_per_second_upper_estimate": 0.0037853269424968938
     },
     {
       "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": 18958.466969986555,
-      "required_pages_per_second_upper_estimate": 0.004385707675883541
+      "required_token_positions_per_second": 18978.115158902427,
+      "required_pages_per_second_upper_estimate": 0.004390252938592913
     },
     {
       "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": 28437.70045497983,
-      "required_pages_per_second_upper_estimate": 0.0017317067566248517
+      "required_token_positions_per_second": 28467.17273835364,
+      "required_pages_per_second_upper_estimate": 0.0017335014640531964
     },
     {
       "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": 28437.70045497983,
-      "required_pages_per_second_upper_estimate": 0.0021892129016175487
+      "required_token_positions_per_second": 28467.17273835364,
+      "required_pages_per_second_upper_estimate": 0.002191481759576167
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28437.70045497983,
-      "required_pages_per_second_upper_estimate": 0.35310106478984604
+      "required_token_positions_per_second": 28467.17273835364,
+      "required_pages_per_second_upper_estimate": 0.3534670119119615
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10091643480458856,
+    "producer_batch_seconds": 0.09455374124217997,
     "producer_p95_seconds": 0.3116126649983926,
-    "gpu_update_seconds": 0.9722305094172429,
-    "producer_batches_per_second": 9.909188745484013,
-    "gpu_batches_per_second": 1.028562661132083,
-    "producer_headroom": 9.63401562193353,
+    "gpu_update_seconds": 0.9712239516764524,
+    "producer_batches_per_second": 10.575996114619151,
+    "gpu_batches_per_second": 1.0296286436036473,
+    "producer_headroom": 10.27166073935522,
     "sustainable": true,
-    "buffer_seconds": 60.278291583869056,
+    "buffer_seconds": 60.215885003940045,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 509421.04264598043,
-    "upload_bytes_per_second": 3979951.465768511,
+    "generation_bytes_per_second": 509643.2155412006,
+    "upload_bytes_per_second": 4008350.9631803427,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 109.30989703688202,
+    "estimated_full_shards_per_hour": 109.35757016827596,
     "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": 1.024016710000069,
-      "bytes": 910901,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 0.949438960000407,
       "bytes": 6133978,
@@ -562,6 +556,12 @@
       "bytes": 7462357,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.5228881869988982,
+      "bytes": 5895199,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -212,3 +212,4 @@
 {"step": 136200, "loss": 0.6191393472254276, "lr": 0.0003, "grad_norm": 0.13696342735562392, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19932615125, "tokens_per_second": 149543.99051949452, "data_wait_s": 1.565600177855231e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 136300, "loss": 0.6153386607766151, "lr": 0.0003, "grad_norm": 0.12918039986370972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19947345253, "tokens_per_second": 150388.36224613513, "data_wait_s": 9.712999599287286e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 136400, "loss": 0.6201961506158113, "lr": 0.0003, "grad_norm": 0.13811877783067825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19962043205, "tokens_per_second": 149190.91475862943, "data_wait_s": 1.1001000530086458e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 136500, "loss": 0.548358827829361, "lr": 0.0003, "grad_norm": 0.15211027111072187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19976590796, "tokens_per_second": 148067.31253867495, "data_wait_s": 1.1274998541921377e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bcf8d1e87a8f427b631db25d342fcf4fc3af2e54e4c8c079cc40a81654ae26e6
-size 456549
+oid sha256:4901cc16c06837b8ac21666ab52b9d488e93b0a3e4c2b17fcd05be9d6042188c
+size 458667