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": 74949.96975815145,
-      "required_pages_per_second_upper_estimate": 0.006496473419674905
+      "required_token_positions_per_second": 75592.8191743543,
+      "required_pages_per_second_upper_estimate": 0.006552193977784418
     },
     {
       "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": 9368.746219768931,
-      "required_pages_per_second_upper_estimate": 0.0037373329423045043
+      "required_token_positions_per_second": 9449.102396794287,
+      "required_pages_per_second_upper_estimate": 0.00376938822275183
     },
     {
       "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": 18737.492439537862,
-      "required_pages_per_second_upper_estimate": 0.004334589107283145
+      "required_token_positions_per_second": 18898.204793588575,
+      "required_pages_per_second_upper_estimate": 0.004371767081951949
     },
     {
       "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": 28106.238659306793,
-      "required_pages_per_second_upper_estimate": 0.001711522472314005
+      "required_token_positions_per_second": 28347.30719038286,
+      "required_pages_per_second_upper_estimate": 0.001726202281067703
     },
     {
       "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": 28106.238659306793,
-      "required_pages_per_second_upper_estimate": 0.0021636960550416627
+      "required_token_positions_per_second": 28347.30719038286,
+      "required_pages_per_second_upper_estimate": 0.0021822541778843026
     },
     {
       "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": 28106.238659306793,
-      "required_pages_per_second_upper_estimate": 0.34898541861885585
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28347.30719038286,
+      "required_pages_per_second_upper_estimate": 0.3519786829703473
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.049299035343736364,
-    "producer_p95_seconds": 0.16839327700017748,
-    "gpu_update_seconds": 0.9836961941132932,
-    "producer_batches_per_second": 20.284372564848855,
-    "gpu_batches_per_second": 1.0165740255825662,
-    "producer_headroom": 19.95366009201792,
+    "producer_batch_seconds": 0.0684410804062523,
+    "producer_p95_seconds": 0.18613033600013296,
+    "gpu_update_seconds": 0.9753307365074836,
+    "producer_batches_per_second": 14.611107745000574,
+    "gpu_batches_per_second": 1.0252932288188246,
+    "producer_headroom": 14.250662478121608,
     "sustainable": true,
-    "buffer_seconds": 60.00546784091089,
+    "buffer_seconds": 60.47050566346398,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 530799.5858097532,
-    "upload_bytes_per_second": 5843728.772253798,
+    "generation_bytes_per_second": 534753.2273098361,
+    "upload_bytes_per_second": 5723910.322731022,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.89723473281332,
+    "estimated_full_shards_per_hour": 114.74559416266739,
     "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": [
@@ -376,6 +376,12 @@
       "bytes": 17354859,
       "pages": 32,
       "shards": 5
+    },
+    {
+      "seconds": 0.8934069070000987,
+      "bytes": 967616,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -37,3 +37,4 @@
 {"step": 85600, "loss": 0.6194258686155081, "lr": 0.0003, "grad_norm": 0.13507982240092406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12527092704, "tokens_per_second": 146957.46692111497, "data_wait_s": 9.594999937689863e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 85700, "loss": 0.6051355823874474, "lr": 0.0003, "grad_norm": 0.1315521462717127, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12541770395, "tokens_per_second": 148511.03501026588, "data_wait_s": 1.2501999663072638e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 85800, "loss": 0.5889550969004631, "lr": 0.0003, "grad_norm": 0.16294887114772708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12556535738, "tokens_per_second": 149503.7780028432, "data_wait_s": 9.805999980017077e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 85900, "loss": 0.6246379911899567, "lr": 0.0003, "grad_norm": 0.13538628748801487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12571121491, "tokens_per_second": 147658.76700833597, "data_wait_s": 8.027000149013475e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:68cb2eb9363dbe506cd6a6159952ad7c3730a11d68552db1aa3ce752157dff41
-size 83550
+oid sha256:5f3713fbf8228dcbea3a0c19ef96620742df93cb37d2ac8bfc74d0db95f6db5d
+size 85668