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": 74708.61055118714,
-      "required_pages_per_second_upper_estimate": 0.006475553015334556
+      "required_token_positions_per_second": 74993.05776517294,
+      "required_pages_per_second_upper_estimate": 0.006500208178917972
     },
     {
       "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": 9338.576318898393,
-      "required_pages_per_second_upper_estimate": 0.003725297717767031
+      "required_token_positions_per_second": 9374.132220646617,
+      "required_pages_per_second_upper_estimate": 0.0037394814985825028
     },
     {
       "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": 18677.152637796786,
-      "required_pages_per_second_upper_estimate": 0.004320630556094588
+      "required_token_positions_per_second": 18748.264441293235,
+      "required_pages_per_second_upper_estimate": 0.004337081020308494
     },
     {
       "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": 28015.728956695177,
-      "required_pages_per_second_upper_estimate": 0.0017060109062926676
+      "required_token_positions_per_second": 28122.396661939852,
+      "required_pages_per_second_upper_estimate": 0.0017125064098998703
     },
     {
       "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": 28015.728956695177,
-      "required_pages_per_second_upper_estimate": 0.0021567283675877277
+      "required_token_positions_per_second": 28122.396661939852,
+      "required_pages_per_second_upper_estimate": 0.0021649399428125696
     },
     {
       "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": 28015.728956695177,
-      "required_pages_per_second_upper_estimate": 0.3478615910289082
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28122.396661939852,
+      "required_pages_per_second_upper_estimate": 0.3491860469342023
     }
   ],
   "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": 0.08303451996846434,
-    "producer_p95_seconds": 0.2721742869980517,
-    "gpu_update_seconds": 0.9868741963750581,
-    "producer_batches_per_second": 12.043183971916617,
-    "gpu_batches_per_second": 1.0133003818249124,
-    "producer_headroom": 11.885107504082193,
+    "producer_batch_seconds": 0.20736153696867632,
+    "producer_p95_seconds": 0.3066512090008473,
+    "gpu_update_seconds": 0.983131001683726,
+    "producer_batches_per_second": 4.822495119483312,
+    "gpu_batches_per_second": 1.0171584440805792,
+    "producer_headroom": 4.741144457432509,
     "sustainable": true,
-    "buffer_seconds": 60.19932597887855,
+    "buffer_seconds": 60.95412210439101,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 494022.765644503,
-    "upload_bytes_per_second": 5947747.0696020685,
+    "generation_bytes_per_second": 503127.79358886415,
+    "upload_bytes_per_second": 6103595.5110718,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.00578524590794,
+    "estimated_full_shards_per_hour": 107.9595122885651,
     "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": 0.9955667570011428,
-      "bytes": 43764,
-      "pages": 1,
-      "shards": 1
-    },
     {
       "seconds": 0.9273372809984721,
       "bytes": 2566992,
@@ -562,6 +556,12 @@
       "bytes": 6035027,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.131166884999402,
+      "bytes": 11754261,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -124,3 +124,4 @@
 {"step": 97400, "loss": 0.6127508264034986, "lr": 0.0003, "grad_norm": 0.12496944282964012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14252830089, "tokens_per_second": 149580.05692383242, "data_wait_s": 8.546001481590793e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 97500, "loss": 0.5998757071793079, "lr": 0.0003, "grad_norm": 0.1421152956588826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14267492681, "tokens_per_second": 148811.02533675297, "data_wait_s": 7.196998922154307e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 97600, "loss": 0.7062435783445835, "lr": 0.0003, "grad_norm": 0.16340439951346944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14282107337, "tokens_per_second": 147795.30343363353, "data_wait_s": 1.7122998542618006e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 97700, "loss": 0.6007490325719118, "lr": 0.0003, "grad_norm": 0.13300135242752106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14296808047, "tokens_per_second": 149448.06676770936, "data_wait_s": 1.540999801363796e-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 sha256:280fe04f2c3da2bad48f4f9fddfa0aa8180949b0cb8f5d1b09e755bbdd335d4b
-size 268599
+oid sha256:2511c7b10942005fc29c4003c192d78a581a305c19e35e3d9f9d3d02a1ab2197
+size 270717