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": 73149.03012872905,
-      "required_pages_per_second_upper_estimate": 0.006340372537036332
+      "required_token_positions_per_second": 74649.51648971024,
+      "required_pages_per_second_upper_estimate": 0.006470430891858268
     },
     {
       "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": 9143.628766091131,
-      "required_pages_per_second_upper_estimate": 0.003647530224226556
+      "required_token_positions_per_second": 9331.18956121378,
+      "required_pages_per_second_upper_estimate": 0.0037223510296847694
     },
     {
       "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": 18287.257532182262,
-      "required_pages_per_second_upper_estimate": 0.004230435185330163
+      "required_token_positions_per_second": 18662.37912242756,
+      "required_pages_per_second_upper_estimate": 0.004317212963318907
     },
     {
       "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": 27430.88629827339,
-      "required_pages_per_second_upper_estimate": 0.0016703970568270686
+      "required_token_positions_per_second": 27993.568683641337,
+      "required_pages_per_second_upper_estimate": 0.0017046614619296559
     },
     {
       "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": 27430.88629827339,
-      "required_pages_per_second_upper_estimate": 0.002111705560794323
+      "required_token_positions_per_second": 27993.568683641337,
+      "required_pages_per_second_upper_estimate": 0.0021550224084244812
     },
     {
       "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": 27430.88629827339,
-      "required_pages_per_second_upper_estimate": 0.3405998025537752
+      "required_token_positions_per_second": 27993.568683641337,
+      "required_pages_per_second_upper_estimate": 0.34758643460324246
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.23744970772651186,
-    "producer_p95_seconds": 0.4237744450001628,
-    "gpu_update_seconds": 1.0079149357175627,
-    "producer_batches_per_second": 4.211418112806325,
-    "gpu_batches_per_second": 0.9921472185428745,
-    "producer_headroom": 4.244751216448966,
+    "producer_batch_seconds": 0.12613883863315323,
+    "producer_p95_seconds": 0.4095588280033553,
+    "gpu_update_seconds": 0.9876554258748983,
+    "producer_batches_per_second": 7.927772372379912,
+    "gpu_batches_per_second": 1.0124988673192035,
+    "producer_headroom": 7.829907398682134,
     "sustainable": true,
-    "buffer_seconds": 60.47489614305376,
+    "buffer_seconds": 60.246980978368796,
     "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": 500576.047084266,
-    "upload_bytes_per_second": 4479140.786293985,
+    "generation_bytes_per_second": 521470.81352168275,
+    "upload_bytes_per_second": 4505439.739794527,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.4119668902968,
+    "estimated_full_shards_per_hour": 111.89549736249792,
     "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": 1.2346246180022717,
-      "bytes": 6315095,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 3.684220068003924,
       "bytes": 2396656,
@@ -562,6 +556,12 @@
       "bytes": 2702936,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.6414927480000188,
+      "bytes": 10577176,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -420,3 +420,4 @@
 {"step": 197000, "loss": 0.6028390862047672, "lr": 0.0003, "grad_norm": 0.1438688130975058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28832906369, "tokens_per_second": 149346.31997957057, "data_wait_s": 2.1495005057659e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 197100, "loss": 0.6232964936643839, "lr": 0.0003, "grad_norm": 0.15152785592092718, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28847548768, "tokens_per_second": 147647.33111312086, "data_wait_s": 9.79300239123404e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 197200, "loss": 0.5884247086942196, "lr": 0.0003, "grad_norm": 0.13629664214417506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28862254588, "tokens_per_second": 150378.736256154, "data_wait_s": 1.0374998964834958e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 197300, "loss": 0.6108404882252216, "lr": 0.0003, "grad_norm": 0.1499395992102569, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28876925716, "tokens_per_second": 149790.0161474689, "data_wait_s": 9.559000318404287e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8c2b497590c8514d854bdb2eae63600eee70c1afb1cb6089e4ccbd60018da721
-size 900225
+oid sha256:05da02563990f121ee6c977997cdd34a4073357fc2231e28b04a22b8e1c9a7cb
+size 902343