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": 75916.54439903462,
-      "required_pages_per_second_upper_estimate": 0.006580253659785628
+      "required_token_positions_per_second": 74315.68651620558,
+      "required_pages_per_second_upper_estimate": 0.006441495355839216
     },
     {
       "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": 9489.568049879328,
-      "required_pages_per_second_upper_estimate": 0.003785530576782882
+      "required_token_positions_per_second": 9289.460814525697,
+      "required_pages_per_second_upper_estimate": 0.003705704808730532
     },
     {
       "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": 18979.136099758656,
-      "required_pages_per_second_upper_estimate": 0.004390489115292056
+      "required_token_positions_per_second": 18578.921629051394,
+      "required_pages_per_second_upper_estimate": 0.004297906541028049
     },
     {
       "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": 28468.704149637982,
-      "required_pages_per_second_upper_estimate": 0.0017335947189657185
+      "required_token_positions_per_second": 27868.38244357709,
+      "required_pages_per_second_upper_estimate": 0.0016970382767111847
     },
     {
       "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": 28468.704149637982,
-      "required_pages_per_second_upper_estimate": 0.0021915996518560533
+      "required_token_positions_per_second": 27868.38244357709,
+      "required_pages_per_second_upper_estimate": 0.0021453852251266435
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28468.704149637982,
-      "required_pages_per_second_upper_estimate": 0.3534860269148091
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27868.38244357709,
+      "required_pages_per_second_upper_estimate": 0.34603204047303837
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09782655885925351,
-    "producer_p95_seconds": 0.31791519399848767,
-    "gpu_update_seconds": 0.9711717068214916,
-    "producer_batches_per_second": 10.222172911537601,
-    "gpu_batches_per_second": 1.0296840331900312,
-    "producer_headroom": 9.927485113922389,
+    "producer_batch_seconds": 0.2398038003830152,
+    "producer_p95_seconds": 0.30634939200172084,
+    "gpu_update_seconds": 0.9920920260074914,
+    "producer_batches_per_second": 4.170075696893868,
+    "gpu_batches_per_second": 1.0079710085205835,
+    "producer_headroom": 4.137098846736039,
     "sustainable": true,
-    "buffer_seconds": 60.21264582293248,
+    "buffer_seconds": 60.517613586456974,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 404336.58502962545,
-    "upload_bytes_per_second": 5256177.876261954,
+    "generation_bytes_per_second": 411473.90439870243,
+    "upload_bytes_per_second": 5223183.501047112,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.76121867338726,
+    "estimated_full_shards_per_hour": 88.29272126169973,
     "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": 0.7203741030007222,
-      "bytes": 1052913,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.0976181589994667,
       "bytes": 7695785,
@@ -562,6 +556,12 @@
       "bytes": 2712693,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 3.9525413480005227,
+      "bytes": 15333188,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -213,3 +213,4 @@
 {"step": 61300, "loss": 0.6554252617061138, "lr": 0.0003, "grad_norm": 0.14056016071280364, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8971511367, "tokens_per_second": 149383.62697999607, "data_wait_s": 1.0161998943658546e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 61400, "loss": 0.6789006367325783, "lr": 0.0003, "grad_norm": 0.15025199092677508, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8986351874, "tokens_per_second": 151812.90684847214, "data_wait_s": 9.141000191448256e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 61500, "loss": 0.6323178894817829, "lr": 0.0003, "grad_norm": 0.1411816376063164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9001065635, "tokens_per_second": 149132.58962462685, "data_wait_s": 1.1607000487856567e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 61600, "loss": 0.5627754870802164, "lr": 0.0003, "grad_norm": 0.13681460468315645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9015734789, "tokens_per_second": 150014.01704744526, "data_wait_s": 1.049899947247468e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:02b17471e49654bbf9ec2451c59d9c2285f493bbc2844c66497c5c04760b2491
-size 458667
+oid sha256:f0654f0558e2040a136cfd49e3f06a4b2374e8de3523586f29b340f4dd48c7f2
+size 460785