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": 74676.27376634613,
-      "required_pages_per_second_upper_estimate": 0.006472750144781369
+      "required_token_positions_per_second": 75681.78761690138,
+      "required_pages_per_second_upper_estimate": 0.006559905536895943
     },
     {
       "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": 9334.534220793266,
-      "required_pages_per_second_upper_estimate": 0.0037236852643981436
+      "required_token_positions_per_second": 9460.223452112672,
+      "required_pages_per_second_upper_estimate": 0.003773824577992928
     },
     {
       "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": 18669.068441586533,
-      "required_pages_per_second_upper_estimate": 0.00431876041957834
+      "required_token_positions_per_second": 18920.446904225344,
+      "required_pages_per_second_upper_estimate": 0.004376912402799988
     },
     {
       "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": 28003.6026623798,
-      "required_pages_per_second_upper_estimate": 0.0017052724785906104
+      "required_token_positions_per_second": 28380.670356338018,
+      "required_pages_per_second_upper_estimate": 0.0017282339228313684
     },
     {
       "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": 28003.6026623798,
-      "required_pages_per_second_upper_estimate": 0.0021557948518835988
+      "required_token_positions_per_second": 28380.670356338018,
+      "required_pages_per_second_upper_estimate": 0.002184822566754685
     },
     {
       "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": 28003.6026623798,
-      "required_pages_per_second_upper_estimate": 0.34771102303760754
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28380.670356338018,
+      "required_pages_per_second_upper_estimate": 0.35239294183217673
     }
   ],
   "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": 0.11668934782028373,
-    "producer_p95_seconds": 0.37897779799823184,
-    "gpu_update_seconds": 0.9873015387817398,
-    "producer_batches_per_second": 8.569762524854674,
-    "gpu_batches_per_second": 1.0128617861103804,
-    "producer_headroom": 8.460939727783108,
+    "producer_batch_seconds": 0.1251538520076565,
+    "producer_p95_seconds": 0.3769494529988151,
+    "gpu_update_seconds": 0.9741841772185484,
+    "producer_batches_per_second": 7.990165575877148,
+    "gpu_batches_per_second": 1.0264999405504203,
+    "producer_headroom": 7.783892877375848,
     "sustainable": true,
-    "buffer_seconds": 60.22539386568612,
+    "buffer_seconds": 60.39941898755001,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 430789.46280473354,
-    "upload_bytes_per_second": 5879434.878421239,
+    "generation_bytes_per_second": 429635.7499050022,
+    "upload_bytes_per_second": 5936969.930406086,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.43739045244698,
+    "estimated_full_shards_per_hour": 92.1898305212264,
     "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": 0.8061036489998514,
-      "bytes": 878150,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.0866164300023229,
       "bytes": 7550162,
@@ -562,6 +556,12 @@
       "bytes": 4594114,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.2605919760026154,
+      "bytes": 7665641,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -310,3 +310,4 @@
 {"step": 116000, "loss": 0.6011932659894228, "lr": 0.0003, "grad_norm": 0.1298500720331202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16975010873, "tokens_per_second": 148774.85320683385, "data_wait_s": 1.6281002899631858e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 116100, "loss": 0.5674298200756311, "lr": 0.0003, "grad_norm": 0.14025242568622506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16989562756, "tokens_per_second": 148254.01830830792, "data_wait_s": 8.544993761461228e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 116200, "loss": 0.63331014290452, "lr": 0.0003, "grad_norm": 0.14362439571118016, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17004101407, "tokens_per_second": 148138.27393720526, "data_wait_s": 1.3131000741850585e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 116300, "loss": 0.6128046456724405, "lr": 0.0003, "grad_norm": 0.15431430011580516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17018720502, "tokens_per_second": 148930.9296571331, "data_wait_s": 2.039800165221095e-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 sha256:37800f58e1ec0431516cd2ad255b7193c48df824f460ced4f51273ce29e66806
-size 665679
+oid sha256:7a78dcaf3029a4bfd4f624554d8818cdf5d2a541995f9930d23d9ac6683d265f
+size 667797