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": 74856.47907658899,
-      "required_pages_per_second_upper_estimate": 0.006488369884347037
+      "required_token_positions_per_second": 75238.60688025369,
+      "required_pages_per_second_upper_estimate": 0.006521491753874629
     },
     {
       "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": 9357.059884573624,
-      "required_pages_per_second_upper_estimate": 0.003732671088468814
+      "required_token_positions_per_second": 9404.825860031711,
+      "required_pages_per_second_upper_estimate": 0.003751725650244021
     },
     {
       "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": 18714.119769147248,
-      "required_pages_per_second_upper_estimate": 0.004329182251333226
+      "required_token_positions_per_second": 18809.651720063423,
+      "required_pages_per_second_upper_estimate": 0.004351281886872771
     },
     {
       "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": 28071.179653720872,
-      "required_pages_per_second_upper_estimate": 0.0017093875628142092
+      "required_token_positions_per_second": 28214.477580095132,
+      "required_pages_per_second_upper_estimate": 0.001718113654704286
     },
     {
       "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": 28071.179653720872,
-      "required_pages_per_second_upper_estimate": 0.0021609971157421195
+      "required_token_positions_per_second": 28214.477580095132,
+      "required_pages_per_second_upper_estimate": 0.0021720285867884584
     },
     {
       "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": 28071.179653720872,
-      "required_pages_per_second_upper_estimate": 0.34855010310442247
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28214.477580095132,
+      "required_pages_per_second_upper_estimate": 0.35032938376268213
     }
   ],
   "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": 0.10251807005465707,
-    "producer_p95_seconds": 0.3422066929997527,
-    "gpu_update_seconds": 0.9849247641552257,
-    "producer_batches_per_second": 9.75437793031857,
-    "gpu_batches_per_second": 1.01530597705877,
-    "producer_headroom": 9.607328382499956,
+    "producer_batch_seconds": 0.2749245478985358,
+    "producer_p95_seconds": 0.3486719409993384,
+    "gpu_update_seconds": 0.9799224501503876,
+    "producer_batches_per_second": 3.6373616239211275,
+    "gpu_batches_per_second": 1.0204889171041354,
+    "producer_headroom": 3.564332314595784,
     "sustainable": true,
-    "buffer_seconds": 60.080410613468764,
+    "buffer_seconds": 60.75519190932403,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384043.67434570723,
-    "upload_bytes_per_second": 4150320.221073577,
+    "generation_bytes_per_second": 382143.40680494235,
+    "upload_bytes_per_second": 4064991.715823895,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.40683243540204,
+    "estimated_full_shards_per_hour": 81.99907925711825,
     "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": 1.218643701999099,
-      "bytes": 11866314,
-      "pages": 38,
-      "shards": 5
-    },
     {
       "seconds": 1.0883352769997146,
       "bytes": 6129558,
@@ -562,6 +556,12 @@
       "bytes": 5847296,
       "pages": 28,
       "shards": 3
+    },
+    {
+      "seconds": 1.0819334130028437,
+      "bytes": 2586941,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -219,3 +219,4 @@
 {"step": 106900, "loss": 0.6316846124827862, "lr": 0.0003, "grad_norm": 0.1368452450334674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15643086303, "tokens_per_second": 147602.88098039146, "data_wait_s": 1.6486003005411476e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 107000, "loss": 0.6474979519844055, "lr": 0.0003, "grad_norm": 0.13514128084320207, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15657524508, "tokens_per_second": 146448.4214527384, "data_wait_s": 1.5558998711640015e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 107100, "loss": 0.5815568156540394, "lr": 0.0003, "grad_norm": 0.14148197939184085, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15672075799, "tokens_per_second": 147734.25784706324, "data_wait_s": 1.8431001080898568e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 107200, "loss": 0.597231900319457, "lr": 0.0003, "grad_norm": 0.13108384581411894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15686710105, "tokens_per_second": 148101.21162240565, "data_wait_s": 1.497999983257614e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b29d26679b191f57241d06c0b7e4aba59a9387ec18d561d6ba48c4ab514a5b78
-size 471375
+oid sha256:d011b5cf9cc46676aac0e9cd51f1e128ecd40f22884495c51477e8c035fbfbfc
+size 473493