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": 74798.79882126283,
-      "required_pages_per_second_upper_estimate": 0.006483370306004635
+      "required_token_positions_per_second": 76035.76043200938,
+      "required_pages_per_second_upper_estimate": 0.006590587003373607
     },
     {
       "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": 9349.849852657853,
-      "required_pages_per_second_upper_estimate": 0.0037297948989380297
+      "required_token_positions_per_second": 9504.470054001173,
+      "required_pages_per_second_upper_estimate": 0.0037914752090319023
     },
     {
       "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": 18699.699705315707,
-      "required_pages_per_second_upper_estimate": 0.0043258464233502545
+      "required_token_positions_per_second": 19008.940108002345,
+      "required_pages_per_second_upper_estimate": 0.004397383748066604
     },
     {
       "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": 28049.549557973558,
-      "required_pages_per_second_upper_estimate": 0.0017080704034675397
+      "required_token_positions_per_second": 28513.410162003518,
+      "required_pages_per_second_upper_estimate": 0.00173631708056446
     },
     {
       "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": 28049.549557973558,
-      "required_pages_per_second_upper_estimate": 0.002159331971095554
+      "required_token_positions_per_second": 28513.410162003518,
+      "required_pages_per_second_upper_estimate": 0.002195041244442115
     },
     {
       "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": 28049.549557973558,
-      "required_pages_per_second_upper_estimate": 0.3482815297065145
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28513.410162003518,
+      "required_pages_per_second_upper_estimate": 0.35404112596699056
     }
   ],
   "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": 0.16826386388305536,
-    "producer_p95_seconds": 0.3738850650006498,
-    "gpu_update_seconds": 0.9856842778475418,
-    "producer_batches_per_second": 5.9430466941791344,
-    "gpu_batches_per_second": 1.0145236385262426,
-    "producer_headroom": 5.8579676889661805,
+    "producer_batch_seconds": 0.1883930876329316,
+    "producer_p95_seconds": 0.3850530769996112,
+    "gpu_update_seconds": 0.9696490122687342,
+    "producer_batches_per_second": 5.308050377880199,
+    "gpu_batches_per_second": 1.0313010041233912,
+    "producer_headroom": 5.146945805984216,
     "sustainable": true,
-    "buffer_seconds": 60.126740948700046,
+    "buffer_seconds": 60.11823876066152,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 289980.40497535205,
-    "upload_bytes_per_second": 5335976.060886515,
+    "generation_bytes_per_second": 290958.70411370526,
+    "upload_bytes_per_second": 5398117.726195879,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.22304450936719,
+    "estimated_full_shards_per_hour": 62.4329647308194,
     "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": 1.1189806990005309,
-      "bytes": 7079866,
-      "pages": 44,
-      "shards": 2
-    },
     {
       "seconds": 1.104166561999591,
       "bytes": 7255454,
@@ -562,6 +556,12 @@
       "bytes": 5422811,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.1521110989997396,
+      "bytes": 12146138,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -282,3 +282,4 @@
 {"step": 143200, "loss": 0.6253639459609985, "lr": 0.0003, "grad_norm": 0.15311304704288073, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20957051223, "tokens_per_second": 147355.76498266353, "data_wait_s": 1.007100217975676e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 143300, "loss": 0.5738432295620441, "lr": 0.0003, "grad_norm": 0.13448258937396648, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20971727827, "tokens_per_second": 149933.2858334302, "data_wait_s": 1.1698997695930302e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 143400, "loss": 0.608228113502264, "lr": 0.0003, "grad_norm": 0.1414706204106599, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20986451027, "tokens_per_second": 150493.9650462964, "data_wait_s": 1.326200072071515e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 143500, "loss": 0.592973381280899, "lr": 0.0003, "grad_norm": 0.14191737250964032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21001089296, "tokens_per_second": 149698.6787272301, "data_wait_s": 1.6450001567136496e-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:b4b87fa0d184c3239ac9055479eaac3997ec7a73cd90747e0cb04099c84b7cda
-size 605592
+oid sha256:5719953593c58f0be4b91524886dd4a43dee1179ae021f024dee593b6faa6c8c
+size 607710