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": 76592.39120717594,
-      "required_pages_per_second_upper_estimate": 0.005298839342544909
+      "required_token_positions_per_second": 77086.5346008687,
+      "required_pages_per_second_upper_estimate": 0.005333025329091214
     },
     {
       "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": 9574.048900896993,
-      "required_pages_per_second_upper_estimate": 0.00400845434592985
+      "required_token_positions_per_second": 9635.816825108588,
+      "required_pages_per_second_upper_estimate": 0.004034315285936312
     },
     {
       "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": 19148.097801793985,
-      "required_pages_per_second_upper_estimate": 0.00398804380281694
+      "required_token_positions_per_second": 19271.633650217176,
+      "required_pages_per_second_upper_estimate": 0.004013773062184086
     },
     {
       "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": 28722.146702690978,
-      "required_pages_per_second_upper_estimate": 0.0017218488386476641
+      "required_token_positions_per_second": 28907.450475325764,
+      "required_pages_per_second_upper_estimate": 0.0017329575168746419
     },
     {
       "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": 28722.146702690978,
-      "required_pages_per_second_upper_estimate": 0.002233020570762412
+      "required_token_positions_per_second": 28907.450475325764,
+      "required_pages_per_second_upper_estimate": 0.0022474271240196094
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28722.146702690978,
-      "required_pages_per_second_upper_estimate": 0.36841685846373157
+      "required_token_positions_per_second": 28907.450475325764,
+      "required_pages_per_second_upper_estimate": 0.3707937362954011
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.3192339207812296,
-    "producer_p95_seconds": 0.2504369569996925,
-    "gpu_update_seconds": 0.9626021441290689,
-    "producer_batches_per_second": 3.1324991954263472,
-    "gpu_batches_per_second": 1.0388507921980243,
-    "producer_headroom": 3.0153504419999853,
+    "producer_batch_seconds": 0.07753917492189544,
+    "producer_p95_seconds": 0.20011221399909118,
+    "gpu_update_seconds": 0.9564316307866451,
+    "producer_batches_per_second": 12.89670674220214,
+    "gpu_batches_per_second": 1.0455530409189007,
+    "producer_headroom": 12.334818261221514,
     "sustainable": true,
-    "buffer_seconds": 60.643935080131335,
+    "buffer_seconds": 60.25519273955864,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346159.08685007953,
-    "upload_bytes_per_second": 5419366.964253601,
+    "generation_bytes_per_second": 332163.06408051273,
+    "upload_bytes_per_second": 5411894.228719663,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.2776818669013,
+    "estimated_full_shards_per_hour": 71.27446119128739,
     "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": 1.1961185629998,
-      "bytes": 11182870,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.2023984399993424,
       "bytes": 8470279,
@@ -562,6 +556,12 @@
       "bytes": 2204351,
       "pages": 24,
       "shards": 2
+    },
+    {
+      "seconds": 1.1051343430008274,
+      "bytes": 10092152,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -128,3 +128,4 @@
 {"step": 569900, "loss": 0.5642459392547607, "lr": 0.0003, "grad_norm": 0.16307487692380424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83416599645, "tokens_per_second": 153346.44949859183, "data_wait_s": 8.794000677880831e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 570000, "loss": 0.5611103773117065, "lr": 0.0003, "grad_norm": 0.16120090148798766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83431290043, "tokens_per_second": 152762.5775838936, "data_wait_s": 8.339000487467274e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 570100, "loss": 0.5936891436576843, "lr": 0.0003, "grad_norm": 0.1690868053704629, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83445911747, "tokens_per_second": 135515.5591265896, "data_wait_s": 2.7539999791770242e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 570200, "loss": 0.587658166885376, "lr": 0.0003, "grad_norm": 0.1683831684998894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83460408739, "tokens_per_second": 151471.25496444936, "data_wait_s": 1.2197000614833087e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:173792e9ba9296e6563f32a73e6da6e16f8da92f72c45bbccd1b9c3ad3f62cf3
-size 277854
+oid sha256:9cb2ca5778fda56feb17a1a8e03ef2909c8afa031b5677c12c0929490dab3849
+size 279972