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": 77332.63672561655,
-      "required_pages_per_second_upper_estimate": 0.005350051245116874
+      "required_token_positions_per_second": 77057.23564710867,
+      "required_pages_per_second_upper_estimate": 0.005330998359487156
     },
     {
       "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": 9666.57959070207,
-      "required_pages_per_second_upper_estimate": 0.004047195013490708
+      "required_token_positions_per_second": 9632.154455888583,
+      "required_pages_per_second_upper_estimate": 0.004032781928422862
     },
     {
       "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": 19333.15918140414,
-      "required_pages_per_second_upper_estimate": 0.0040265872077929615
+      "required_token_positions_per_second": 19264.308911777167,
+      "required_pages_per_second_upper_estimate": 0.0040122475123333745
     },
     {
       "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": 28999.738772106208,
-      "required_pages_per_second_upper_estimate": 0.0017384900593505595
+      "required_token_positions_per_second": 28896.463367665747,
+      "required_pages_per_second_upper_estimate": 0.0017322988565467658
     },
     {
       "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": 28999.738772106208,
-      "required_pages_per_second_upper_estimate": 0.002254602133160972
+      "required_token_positions_per_second": 28896.463367665747,
+      "required_pages_per_second_upper_estimate": 0.0022465729247262223
     },
     {
       "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": 28999.738772106208,
-      "required_pages_per_second_upper_estimate": 0.3719775114750479
+      "required_token_positions_per_second": 28896.463367665747,
+      "required_pages_per_second_upper_estimate": 0.3706528054753755
     }
   ],
   "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": 0.04990577521873618,
-    "producer_p95_seconds": 0.15678710700012743,
-    "gpu_update_seconds": 0.9533878983280223,
-    "producer_batches_per_second": 20.037761073082557,
-    "gpu_batches_per_second": 1.04889101461611,
-    "producer_headroom": 19.103758916665235,
+    "producer_batch_seconds": 0.28105300232029506,
+    "producer_p95_seconds": 0.16417822099992918,
+    "gpu_update_seconds": 0.9567952883444297,
+    "producer_batches_per_second": 3.5580477409751166,
+    "gpu_batches_per_second": 1.0451556484254105,
+    "producer_headroom": 3.4043233142695333,
     "sustainable": true,
-    "buffer_seconds": 60.0634375946654,
+    "buffer_seconds": 60.27810316569907,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 421557.4367991584,
-    "upload_bytes_per_second": 5049947.811691667,
+    "generation_bytes_per_second": 402230.7536586422,
+    "upload_bytes_per_second": 5248915.094982789,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.45641258221687,
+    "estimated_full_shards_per_hour": 86.30935628241967,
     "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": [
@@ -268,6 +268,12 @@
       "bytes": 11919197,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.3387522560001344,
+      "bytes": 10892645,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -30,3 +30,4 @@
 {"step": 888000, "loss": 0.49947598576545715, "lr": 3e-05, "grad_norm": 0.3130021412299721, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129978566661, "tokens_per_second": 151972.8083974794, "data_wait_s": 1.097999984267517e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 888100, "loss": 0.49853742122650146, "lr": 3e-05, "grad_norm": 0.2948677189220878, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129993158515, "tokens_per_second": 152198.22789579048, "data_wait_s": 1.2349999906291487e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 888200, "loss": 0.5294581055641174, "lr": 3e-05, "grad_norm": 0.3253916632741411, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130007913918, "tokens_per_second": 153360.87377444454, "data_wait_s": 1.2444999811123125e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 888300, "loss": 0.5584107637405396, "lr": 3e-05, "grad_norm": 0.33135345481047074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130022586638, "tokens_per_second": 153200.12345570748, "data_wait_s": 1.2900000001536682e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3fa496ffb9334645ffd42c5a62a28e694dfde6b61e00579507ea514991fbe198
-size 67941
+oid sha256:43924345bf6de502ddabd40ad4c87003c0e2d75aca36fbefee4485f5a46935b9
+size 70059