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": 77262.40200108931,
-      "required_pages_per_second_upper_estimate": 0.005345192243907066
+      "required_token_positions_per_second": 76657.15316727746,
+      "required_pages_per_second_upper_estimate": 0.005303319725213183
     },
     {
       "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": 9657.800250136164,
-      "required_pages_per_second_upper_estimate": 0.004043519286929241
+      "required_token_positions_per_second": 9582.144145909682,
+      "required_pages_per_second_upper_estimate": 0.004011843655968724
     },
     {
       "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": 19315.60050027233,
-      "required_pages_per_second_upper_estimate": 0.004022930197566766
+      "required_token_positions_per_second": 19164.288291819365,
+      "required_pages_per_second_upper_estimate": 0.003991415854917291
     },
     {
       "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": 28973.40075040849,
-      "required_pages_per_second_upper_estimate": 0.0017369111351656118
+      "required_token_positions_per_second": 28746.432437729047,
+      "required_pages_per_second_upper_estimate": 0.0017233047313810262
     },
     {
       "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": 28973.40075040849,
-      "required_pages_per_second_upper_estimate": 0.0022525544678226885
+      "required_token_positions_per_second": 28746.432437729047,
+      "required_pages_per_second_upper_estimate": 0.002234908679839963
     },
     {
       "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": 28973.40075040849,
-      "required_pages_per_second_upper_estimate": 0.37163967561227396
+      "required_token_positions_per_second": 28746.432437729047,
+      "required_pages_per_second_upper_estimate": 0.3687283697968093
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.050401381210949836,
-    "producer_p95_seconds": 0.16159123900024497,
-    "gpu_update_seconds": 0.9542545674280295,
-    "producer_batches_per_second": 19.840726106584302,
-    "gpu_batches_per_second": 1.0479383951970664,
-    "producer_headroom": 18.933103508296615,
+    "producer_batch_seconds": 0.07536875569529755,
+    "producer_p95_seconds": 0.16106941199996072,
+    "gpu_update_seconds": 0.961788912759575,
+    "producer_batches_per_second": 13.268097513017487,
+    "gpu_batches_per_second": 1.0397291824988806,
+    "producer_headroom": 12.76110908143311,
     "sustainable": true,
-    "buffer_seconds": 60.11803774796586,
+    "buffer_seconds": 60.59270150385323,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330426.1723661001,
+    "generation_bytes_per_second": 326067.8161344017,
     "upload_bytes_per_second": 5609063.499692655,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.9017646621442,
+    "estimated_full_shards_per_hour": 69.96656287216223,
     "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": [

metrics/void-byte-v1/1791033297667653750-ba4975eb/train.jsonl

@@ -32,3 +32,4 @@
 {"step": 603200, "loss": 0.6052895784378052, "lr": 0.0003, "grad_norm": 0.1582190484922652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88290369731, "tokens_per_second": 151266.30136291697, "data_wait_s": 1.689000009719166e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 603300, "loss": 0.5545015931129456, "lr": 0.0003, "grad_norm": 0.1697075137537699, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88304881610, "tokens_per_second": 151125.20079138578, "data_wait_s": 1.1192999863851583e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 603400, "loss": 0.5984642505645752, "lr": 0.0003, "grad_norm": 0.17757429015141551, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88319406939, "tokens_per_second": 150195.53006902183, "data_wait_s": 1.5753999832668342e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 603500, "loss": 0.6747608184814453, "lr": 0.0003, "grad_norm": 0.2245971242190504, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88334247204, "tokens_per_second": 147668.0634139631, "data_wait_s": 1.7825999748311006e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791033297.83ac8725-d818-41af-bbb1-78dcd88792cf-7zwhs.737.0.1791033297667653750-ba4975eb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a1b4eb914fce3b251e3843aba8b4ca5140153aa5a2a2e24fc97886c35bb6fb41
-size 72177
+oid sha256:262f3737a576710cc1a17f0aac1daa50345a18d3d949f6ce9748f07daaf1e535
+size 74295