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": 77484.03549020716,
-      "required_pages_per_second_upper_estimate": 0.005360525362944784
+      "required_token_positions_per_second": 77574.96884525687,
+      "required_pages_per_second_upper_estimate": 0.005366816343441576
     },
     {
       "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": 9685.504436275894,
-      "required_pages_per_second_upper_estimate": 0.004055118451136754
+      "required_token_positions_per_second": 9696.871105657108,
+      "required_pages_per_second_upper_estimate": 0.004059877438243619
     },
     {
       "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": 19371.00887255179,
-      "required_pages_per_second_upper_estimate": 0.004034470300295538
+      "required_token_positions_per_second": 19393.742211314217,
+      "required_pages_per_second_upper_estimate": 0.004039205055241293
     },
     {
       "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": 29056.51330882768,
-      "required_pages_per_second_upper_estimate": 0.0017418936061373137
+      "required_token_positions_per_second": 29090.61331697132,
+      "required_pages_per_second_upper_estimate": 0.0017439378495578274
     },
     {
       "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": 29056.51330882768,
-      "required_pages_per_second_upper_estimate": 0.0022590161036662713
+      "required_token_positions_per_second": 29090.61331697132,
+      "required_pages_per_second_upper_estimate": 0.0022616672293093547
     },
     {
       "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": 29056.51330882768,
-      "required_pages_per_second_upper_estimate": 0.37270575427236285
+      "required_token_positions_per_second": 29090.61331697132,
+      "required_pages_per_second_upper_estimate": 0.3731431525630934
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.047726385570324936,
-    "producer_p95_seconds": 0.1562362069998926,
-    "gpu_update_seconds": 0.9515250403977493,
-    "producer_batches_per_second": 20.952770423532236,
-    "gpu_batches_per_second": 1.0509444917834085,
-    "producer_headroom": 19.937085723696278,
+    "producer_batch_seconds": 0.07163538224216381,
+    "producer_p95_seconds": 0.1543780169999991,
+    "gpu_update_seconds": 0.9504096630327932,
+    "producer_batches_per_second": 13.959582104545689,
+    "gpu_batches_per_second": 1.0521778543464744,
+    "producer_headroom": 13.267321724059878,
     "sustainable": true,
-    "buffer_seconds": 60.89760258545596,
+    "buffer_seconds": 60.82621843409876,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 423444.3501771704,
+    "generation_bytes_per_second": 421845.64496211434,
     "upload_bytes_per_second": 5055656.316842427,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.8613002680429,
+    "estimated_full_shards_per_hour": 90.51825534484456,
     "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-v1/1791399463488568880-054ba646/train.jsonl

@@ -32,3 +32,4 @@
 {"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}
 {"step": 888400, "loss": 0.5353817343711853, "lr": 3e-05, "grad_norm": 0.3351649442302755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130037293845, "tokens_per_second": 153090.013600368, "data_wait_s": 1.27170001178456e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 888500, "loss": 0.537222683429718, "lr": 3e-05, "grad_norm": 0.3170551465203647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130051959022, "tokens_per_second": 153041.954718379, "data_wait_s": 1.0757999916677363e-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 sha256:31edc450ab0d0624bf7bd58bc4b8d539fa8b455e0276ba1731f02eb233564dbf
-size 72177
+oid sha256:42edc82545be03c49d8b0f68deabf476850cc027e39e4cdbc64950775986abd6
+size 74295