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": 74822.26596480944,
-      "required_pages_per_second_upper_estimate": 0.0064854043785303244
+      "required_token_positions_per_second": 75280.79771540457,
+      "required_pages_per_second_upper_estimate": 0.006525148748533813
     },
     {
       "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": 9352.78324560118,
-      "required_pages_per_second_upper_estimate": 0.0037309650732412557
+      "required_token_positions_per_second": 9410.09971442557,
+      "required_pages_per_second_upper_estimate": 0.0037538294696128813
     },
     {
       "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": 18705.56649120236,
-      "required_pages_per_second_upper_estimate": 0.004327203601013227
+      "required_token_positions_per_second": 18820.19942885114,
+      "required_pages_per_second_upper_estimate": 0.004353721913667475
     },
     {
       "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": 28058.349736803542,
-      "required_pages_per_second_upper_estimate": 0.0017086062881873155
+      "required_token_positions_per_second": 28230.299143276712,
+      "required_pages_per_second_upper_estimate": 0.0017190771048927176
     },
     {
       "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": 28058.349736803542,
-      "required_pages_per_second_upper_estimate": 0.0021600094332223404
+      "required_token_positions_per_second": 28230.299143276712,
+      "required_pages_per_second_upper_estimate": 0.0021732465745190666
     },
     {
       "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": 28058.349736803542,
-      "required_pages_per_second_upper_estimate": 0.348390798475279
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28230.299143276712,
+      "required_pages_per_second_upper_estimate": 0.35052583462603165
     }
   ],
   "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": 0.4513572939927144,
-    "producer_p95_seconds": 0.40074575600010576,
-    "gpu_update_seconds": 0.9853751293054382,
-    "producer_batches_per_second": 2.2155396917461614,
-    "gpu_batches_per_second": 1.0148419320313782,
-    "producer_headroom": 2.1831377102357044,
+    "producer_batch_seconds": 0.41631964140646005,
+    "producer_p95_seconds": 0.4022427370000514,
+    "gpu_update_seconds": 0.9793732563611395,
+    "producer_batches_per_second": 2.40200053166284,
+    "gpu_batches_per_second": 1.021061166929858,
+    "producer_headroom": 2.352455082475824,
     "sustainable": true,
-    "buffer_seconds": 60.10788288763173,
+    "buffer_seconds": 60.72114189439065,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 296425.6416624126,
-    "upload_bytes_per_second": 5782082.988884923,
+    "generation_bytes_per_second": 279105.0522586036,
+    "upload_bytes_per_second": 5790253.266318516,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.60604226497921,
+    "estimated_full_shards_per_hour": 59.889446981607264,
     "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": 1.1773676480006543,
-      "bytes": 5012862,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.197479630001908,
       "bytes": 5919458,
@@ -562,6 +556,12 @@
       "bytes": 5948596,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.030731860002561,
+      "bytes": 4760298,
+      "pages": 25,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -390,3 +390,4 @@
 {"step": 234000, "loss": 0.6060496270656586, "lr": 0.0003, "grad_norm": 0.13822100157510223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34251105315, "tokens_per_second": 149677.64584513608, "data_wait_s": 1.2583004718180746e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 234100, "loss": 0.6063950918614864, "lr": 0.0003, "grad_norm": 0.15298467270971142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34265917030, "tokens_per_second": 150056.52678457572, "data_wait_s": 1.572600012877956e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 234200, "loss": 0.6237741932272911, "lr": 0.0003, "grad_norm": 0.1550310669100969, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34280619420, "tokens_per_second": 136297.61284754198, "data_wait_s": 1.2418997357599437e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 234300, "loss": 0.5962309408932924, "lr": 0.0003, "grad_norm": 0.13739073586915035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34295097165, "tokens_per_second": 147068.08691130526, "data_wait_s": 8.19200067780912e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e6fbd11997effa7362ad0cff2a924b53af2754c693c71ca50432e733895846b3
-size 835902
+oid sha256:1906a21fa379fb147dcdd68559d0df8c7b180c3ed8a618914bc7c51d1b94670f
+size 838020