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": 74842.38362670901,
-      "required_pages_per_second_upper_estimate": 0.006487148126476031
+      "required_token_positions_per_second": 74757.05328348005,
+      "required_pages_per_second_upper_estimate": 0.006479751908592723
     },
     {
       "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": 9355.297953338626,
-      "required_pages_per_second_upper_estimate": 0.0037319682277559543
+      "required_token_positions_per_second": 9344.631660435007,
+      "required_pages_per_second_upper_estimate": 0.00372771328404141
     },
     {
       "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": 18710.595906677252,
-      "required_pages_per_second_upper_estimate": 0.004328367067768654
+      "required_token_positions_per_second": 18689.263320870014,
+      "required_pages_per_second_upper_estimate": 0.004323432149482836
     },
     {
       "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": 28065.89386001588,
-      "required_pages_per_second_upper_estimate": 0.0017090656857099915
+      "required_token_positions_per_second": 28033.89498130502,
+      "required_pages_per_second_upper_estimate": 0.001707117121881642
     },
     {
       "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": 28065.89386001588,
-      "required_pages_per_second_upper_estimate": 0.002160590200710696
+      "required_token_positions_per_second": 28033.89498130502,
+      "required_pages_per_second_upper_estimate": 0.002158126838448968
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28065.89386001588,
-      "required_pages_per_second_upper_estimate": 0.34848447123702
+      "required_token_positions_per_second": 28033.89498130502,
+      "required_pages_per_second_upper_estimate": 0.34808715225678905
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10386628711688672,
-    "producer_p95_seconds": 0.4108484979951754,
-    "gpu_update_seconds": 0.9851102600864876,
-    "producer_batches_per_second": 9.627763038016775,
-    "gpu_batches_per_second": 1.0151147952841393,
-    "producer_headroom": 9.484408150431777,
+    "producer_batch_seconds": 0.1454626460472923,
+    "producer_p95_seconds": 0.4284910379938083,
+    "gpu_update_seconds": 0.9862346997603307,
+    "producer_batches_per_second": 6.874617141742929,
+    "gpu_batches_per_second": 1.013957428432618,
+    "producer_headroom": 6.77998597275406,
     "sustainable": true,
-    "buffer_seconds": 60.09172586527574,
+    "buffer_seconds": 60.160316685380174,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 253993.65128372642,
-    "upload_bytes_per_second": 5716823.208343671,
+    "generation_bytes_per_second": 274868.1144998723,
+    "upload_bytes_per_second": 5756299.215187626,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.50112489589543,
+    "estimated_full_shards_per_hour": 58.980298769446634,
     "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": 1.2517391190049239,
-      "bytes": 5921573,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.3183760180036188,
       "bytes": 12970748,
@@ -562,6 +556,12 @@
       "bytes": 11867525,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.0216524100032984,
+      "bytes": 7553152,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -407,3 +407,4 @@
 {"step": 235700, "loss": 0.5529085770249367, "lr": 0.0003, "grad_norm": 0.14916057235719044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34499840230, "tokens_per_second": 148132.54662396782, "data_wait_s": 1.955799962161109e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 235800, "loss": 0.5849977470934391, "lr": 0.0003, "grad_norm": 0.1671746701560514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34514496775, "tokens_per_second": 148640.04987322216, "data_wait_s": 9.06200148165226e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 235900, "loss": 0.601128863170743, "lr": 0.0003, "grad_norm": 0.13912574180998558, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34528941017, "tokens_per_second": 146718.09281255124, "data_wait_s": 7.586997526232153e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 236000, "loss": 0.5956513024866581, "lr": 0.0003, "grad_norm": 0.1497077140744594, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34543587690, "tokens_per_second": 148646.84811544805, "data_wait_s": 1.3183998817112297e-05, "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 sha256:9bae36976ad5ffb0141a900f6833e498b2165667f96cca58143fd669c1c2ad87
-size 872691
+oid sha256:20db44e91d77653965d2c7b322bf601a5eb9386b78ef635712a4540f7ddc46ae
+size 874809