Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75522.8626573945,
-      "required_pages_per_second_upper_estimate": 0.005224846875822889
+      "required_token_positions_per_second": 76710.58553286047,
+      "required_pages_per_second_upper_estimate": 0.005307016300244384
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9440.357832174312,
-      "required_pages_per_second_upper_estimate": 0.0039524806872426425
+      "required_token_positions_per_second": 9588.823191607558,
+      "required_pages_per_second_upper_estimate": 0.004014640032928091
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18880.715664348623,
-      "required_pages_per_second_upper_estimate": 0.003932355154928217
+      "required_token_positions_per_second": 19177.646383215117,
+      "required_pages_per_second_upper_estimate": 0.003994197993078467
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28321.07349652294,
-      "required_pages_per_second_upper_estimate": 0.0016978051123412168
+      "required_token_positions_per_second": 28766.469574822677,
+      "required_pages_per_second_upper_estimate": 0.0017245059271548526
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28321.07349652294,
-      "required_pages_per_second_upper_estimate": 0.0022018388931174413
+      "required_token_positions_per_second": 28766.469574822677,
+      "required_pages_per_second_upper_estimate": 0.0022364664791149408
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28321.07349652294,
-      "required_pages_per_second_upper_estimate": 0.3632723220138651
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28766.469574822677,
+      "required_pages_per_second_upper_estimate": 0.36898538467724473
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07327977911714356,
-    "producer_p95_seconds": 0.266063461000158,
-    "gpu_update_seconds": 0.9762341813559584,
-    "producer_batches_per_second": 13.646329342797559,
-    "gpu_batches_per_second": 1.024344382831414,
-    "producer_headroom": 13.32201315447977,
+    "producer_batch_seconds": 0.0882111206484808,
+    "producer_p95_seconds": 0.2659228590018756,
+    "gpu_update_seconds": 0.9611189836168983,
+    "producer_batches_per_second": 11.336439132033885,
+    "gpu_batches_per_second": 1.0404539053393618,
+    "producer_headroom": 10.89566685641524,
     "sustainable": true,
-    "buffer_seconds": 60.52651924406942,
+    "buffer_seconds": 60.55049596786459,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 393832.94663953065,
-    "upload_bytes_per_second": 5554575.580971991,
+    "generation_bytes_per_second": 393537.5578407053,
+    "upload_bytes_per_second": 5534601.836951126,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.50738238706055,
+    "estimated_full_shards_per_hour": 84.44399882713194,
     "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": 1.1636577470017073,
-      "bytes": 8230638,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.2417163159989286,
       "bytes": 6099161,
@@ -562,6 +556,12 @@
       "bytes": 11418078,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 1.0558311360000516,
+      "bytes": 6077678,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -197,3 +197,4 @@
 {"step": 504700, "loss": 0.5502405166625977, "lr": 0.0003, "grad_norm": 0.16135350636367934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73871415633, "tokens_per_second": 153141.15619234374, "data_wait_s": 1.1763000657083467e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 504800, "loss": 0.6012311577796936, "lr": 0.0003, "grad_norm": 0.1568368892366518, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73886002327, "tokens_per_second": 151655.42577105787, "data_wait_s": 9.138999303104356e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 504900, "loss": 0.5882214903831482, "lr": 0.0003, "grad_norm": 0.15479383777678637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73900679335, "tokens_per_second": 153322.55105851224, "data_wait_s": 8.228998922277242e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 505000, "loss": 0.5756577253341675, "lr": 0.0003, "grad_norm": 0.16870856170823553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73915057124, "tokens_per_second": 149524.0963532161, "data_wait_s": 2.45060000452213e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e562d942b35337c3376d4b811ba4a548aa57b51932af16bdedc639a78c3a32da
-size 423996
+oid sha256:ddfa8e55f15964247c7e40eadcdf9fbbdc28d9eb3275d4719ed4429cb1e5bc3e
+size 426114