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": 72610.239013736,
-      "required_pages_per_second_upper_estimate": 0.006293671488742331
+      "required_token_positions_per_second": 74578.84466500721,
+      "required_pages_per_second_upper_estimate": 0.006464305237209112
     },
     {
       "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": 9076.279876717,
-      "required_pages_per_second_upper_estimate": 0.0036206637452995856
+      "required_token_positions_per_second": 9322.355583125902,
+      "required_pages_per_second_upper_estimate": 0.0037188270237457723
     },
     {
       "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": 18152.559753434,
-      "required_pages_per_second_upper_estimate": 0.004199275224816693
+      "required_token_positions_per_second": 18644.711166251804,
+      "required_pages_per_second_upper_estimate": 0.0043131257926030545
     },
     {
       "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": 27228.839630151,
-      "required_pages_per_second_upper_estimate": 0.0016580934747953571
+      "required_token_positions_per_second": 27967.066749377707,
+      "required_pages_per_second_upper_estimate": 0.001703047633177904
     },
     {
       "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": 27228.839630151,
-      "required_pages_per_second_upper_estimate": 0.002096151448981287
+      "required_token_positions_per_second": 27967.066749377707,
+      "required_pages_per_second_upper_estimate": 0.0021529822161628176
     },
     {
       "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": 27228.839630151,
-      "required_pages_per_second_upper_estimate": 0.3380910591423942
+      "required_token_positions_per_second": 27967.066749377707,
+      "required_pages_per_second_upper_estimate": 0.34725736927595646
     }
   ],
   "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": 60,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.5901036710075971,
-    "producer_p95_seconds": 0.3487042749984539,
-    "gpu_update_seconds": 1.0153939857717937,
-    "producer_batches_per_second": 1.6946174869451471,
-    "gpu_batches_per_second": 0.9848393963451607,
-    "producer_headroom": 1.7207044044278135,
+    "producer_batch_seconds": 0.10460819329665583,
+    "producer_p95_seconds": 0.32829709900033777,
+    "gpu_update_seconds": 0.9885913402275265,
+    "producer_batches_per_second": 9.559480653337777,
+    "gpu_batches_per_second": 1.0115403193495989,
+    "producer_headroom": 9.450419790962304,
     "sustainable": true,
-    "buffer_seconds": 60.92363914630762,
+    "buffer_seconds": 60.304071753879114,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 265919.8301118163,
-    "upload_bytes_per_second": 4475092.595384433,
+    "generation_bytes_per_second": 277735.62911691656,
+    "upload_bytes_per_second": 4560568.163164613,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.06020524516932,
+    "estimated_full_shards_per_hour": 59.595600653940416,
     "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": 2.698638427998958,
-      "bytes": 6167283,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 2.6425060760011547,
       "bytes": 2623268,
@@ -562,6 +556,12 @@
       "bytes": 5704060,
       "pages": 20,
       "shards": 2
+    },
+    {
+      "seconds": 1.1505642880001687,
+      "bytes": 6924827,
+      "pages": 43,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -297,3 +297,4 @@
 {"step": 266800, "loss": 0.5808506812900305, "lr": 0.0003, "grad_norm": 0.14747029966756658, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39051941851, "tokens_per_second": 149506.29244558216, "data_wait_s": 1.3803000911138952e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 266900, "loss": 0.5238931700587273, "lr": 0.0003, "grad_norm": 0.14914426471055756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39066544644, "tokens_per_second": 148746.8750300812, "data_wait_s": 9.159000910585746e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 267000, "loss": 0.579886456951499, "lr": 0.0003, "grad_norm": 0.16076159281022803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39081075379, "tokens_per_second": 146875.19622043442, "data_wait_s": 1.2333002814557403e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 267100, "loss": 0.5743216760456562, "lr": 0.0003, "grad_norm": 0.15949495605024194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39095712282, "tokens_per_second": 148986.0278657728, "data_wait_s": 1.5027999324956909e-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 sha256:56aae9513a6e01657b67d702488c225351e063aa411f76991f55d3438daf039f
-size 638145
+oid sha256:c31b9f85755ee01878a36715f681b2fd76cd195070e753b5fc1ccc161a93db68
+size 640263