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": 73946.04720276408,
-      "required_pages_per_second_upper_estimate": 0.006409455956992382
+      "required_token_positions_per_second": 76131.3579957731,
+      "required_pages_per_second_upper_estimate": 0.006598873157910835
     },
     {
       "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": 9243.25590034551,
-      "required_pages_per_second_upper_estimate": 0.003687272977639026
+      "required_token_positions_per_second": 9516.419749471637,
+      "required_pages_per_second_upper_estimate": 0.0037962421212189394
     },
     {
       "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": 18486.51180069102,
-      "required_pages_per_second_upper_estimate": 0.0042765291535943085
+      "required_token_positions_per_second": 19032.839498943275,
+      "required_pages_per_second_upper_estimate": 0.004402912451545873
     },
     {
       "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": 27729.76770103653,
-      "required_pages_per_second_upper_estimate": 0.0016885973661458133
+      "required_token_positions_per_second": 28549.259248414914,
+      "required_pages_per_second_upper_estimate": 0.0017385001018412937
     },
     {
       "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": 27729.76770103653,
-      "required_pages_per_second_upper_estimate": 0.002134714278535705
+      "required_token_positions_per_second": 28549.259248414914,
+      "required_pages_per_second_upper_estimate": 0.002197801006350687
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 27729.76770103653,
-      "required_pages_per_second_upper_estimate": 0.3443109092843852
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28549.259248414914,
+      "required_pages_per_second_upper_estimate": 0.3544862516410459
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2549577297734231,
-    "producer_p95_seconds": 0.34950604600453516,
-    "gpu_update_seconds": 0.9970512662811282,
-    "producer_batches_per_second": 3.9222187963812045,
-    "gpu_batches_per_second": 1.0029574544645736,
-    "producer_headroom": 3.9106532175635222,
+    "producer_batch_seconds": 0.1340564343671531,
+    "producer_p95_seconds": 0.3527957210026216,
+    "gpu_update_seconds": 0.9684314314226926,
+    "producer_batches_per_second": 7.459544964930255,
+    "gpu_batches_per_second": 1.0325976290659329,
+    "producer_headroom": 7.224057808149346,
     "sustainable": true,
-    "buffer_seconds": 60.82012724314882,
+    "buffer_seconds": 60.04274874820694,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 401570.8620309348,
-    "upload_bytes_per_second": 4667469.659235057,
+    "generation_bytes_per_second": 401458.3315093647,
+    "upload_bytes_per_second": 4637308.27795431,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.16775890060455,
+    "estimated_full_shards_per_hour": 86.1436124702521,
     "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": 1.3881743410020135,
-      "bytes": 10608461,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.1401566910026304,
       "bytes": 8280901,
@@ -562,6 +556,12 @@
       "bytes": 16767391,
       "pages": 42,
       "shards": 4
+    },
+    {
+      "seconds": 1.2854500899993582,
+      "bytes": 7334524,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -327,3 +327,4 @@
 {"step": 187700, "loss": 0.5929309464991093, "lr": 0.0003, "grad_norm": 0.14143837608252244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27471036553, "tokens_per_second": 152747.38312879467, "data_wait_s": 7.114002073649317e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 187800, "loss": 0.6040479615330696, "lr": 0.0003, "grad_norm": 0.14987953626946238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27485684795, "tokens_per_second": 149245.0930823913, "data_wait_s": 7.321999873965979e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 187900, "loss": 0.5844988469034433, "lr": 0.0003, "grad_norm": 0.15019941374902543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27500261351, "tokens_per_second": 149877.07558403688, "data_wait_s": 1.1980999261140823e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 188000, "loss": 0.6003498584032059, "lr": 0.0003, "grad_norm": 0.15873858486695672, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27514881760, "tokens_per_second": 150405.78853102526, "data_wait_s": 9.98800533125177e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:351c3e85e1bd105fc684aae69b16cf771e31068585b43834a10aec29b42d7533
-size 701685
+oid sha256:7dd133210441d33ed680bcdca34cf534ed5191ee691cce94ce99478cd88f06d4
+size 703803