Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75882.46047779212,
-      "required_pages_per_second_upper_estimate": 0.005249724687419895
+      "required_token_positions_per_second": 74525.72483680306,
+      "required_pages_per_second_upper_estimate": 0.005155862567715866
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9485.307559724015,
-      "required_pages_per_second_upper_estimate": 0.003971300199510654
+      "required_token_positions_per_second": 9315.715604600382,
+      "required_pages_per_second_upper_estimate": 0.0039002955893831276
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18970.61511944803,
-      "required_pages_per_second_upper_estimate": 0.003951078840617362
+      "required_token_positions_per_second": 18631.431209200764,
+      "required_pages_per_second_upper_estimate": 0.0038804357769940873
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28455.922679172047,
-      "required_pages_per_second_upper_estimate": 0.0017058891149382511
+      "required_token_positions_per_second": 27947.146813801144,
+      "required_pages_per_second_upper_estimate": 0.0016753887786650306
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28455.922679172047,
-      "required_pages_per_second_upper_estimate": 0.00221232282392954
+      "required_token_positions_per_second": 27947.146813801144,
+      "required_pages_per_second_upper_estimate": 0.002172767738265479
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28455.922679172047,
-      "required_pages_per_second_upper_estimate": 0.3650020225391163
+      "required_token_positions_per_second": 27947.146813801144,
+      "required_pages_per_second_upper_estimate": 0.35847599201910113
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10641431016406955,
-    "producer_p95_seconds": 0.2784876080004324,
-    "gpu_update_seconds": 0.9716079254121887,
-    "producer_batches_per_second": 9.397232369013155,
-    "gpu_batches_per_second": 1.0292217404214425,
-    "producer_headroom": 9.130425446673138,
+    "producer_batch_seconds": 0.09519893699223303,
+    "producer_p95_seconds": 0.27535133800120093,
+    "gpu_update_seconds": 0.9892959801659102,
+    "producer_batches_per_second": 10.504318972401832,
+    "gpu_batches_per_second": 1.0108198355686178,
+    "producer_headroom": 10.391880533777638,
     "sustainable": true,
-    "buffer_seconds": 60.239691375555694,
+    "buffer_seconds": 60.347054790120524,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329945.4941399559,
-    "upload_bytes_per_second": 5176813.427477496,
+    "generation_bytes_per_second": 330116.57259299885,
+    "upload_bytes_per_second": 5233794.275715616,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.79862230443008,
+    "estimated_full_shards_per_hour": 70.83533175795054,
     "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": 3.6534681639986957,
-      "bytes": 12129367,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 0.9705111450002732,
       "bytes": 2305053,
@@ -562,6 +556,12 @@
       "bytes": 10682926,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.9240188730000227,
+      "bytes": 2599575,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -225,3 +225,4 @@
 {"step": 626900, "loss": 0.5559431910514832, "lr": 0.0001, "grad_norm": 0.16029947237022812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91761452094, "tokens_per_second": 151988.67797562748, "data_wait_s": 1.8910999642685056e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 627000, "loss": 0.47953253984451294, "lr": 0.0001, "grad_norm": 0.175877610725448, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91776072408, "tokens_per_second": 151197.93908179356, "data_wait_s": 1.3818000297760591e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 627100, "loss": 0.5235676765441895, "lr": 0.0001, "grad_norm": 0.15805174044136228, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91790634843, "tokens_per_second": 150737.7297824733, "data_wait_s": 1.2975000572623685e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 627200, "loss": 0.5442319512367249, "lr": 0.0001, "grad_norm": 0.1723691441592288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91805318673, "tokens_per_second": 151585.0667025374, "data_wait_s": 1.603900091140531e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:411c0ff93d8a64584423a9985d4fb9e1024a322547b23a7780b8bfc12c0c96b5
-size 484866
+oid sha256:7bf1301ed3d4faa15ee47cf2a5c23d8a147e783604557537babfdc8b54803922
+size 486984