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": 76504.9110552513,
-      "required_pages_per_second_upper_estimate": 0.005292787262652845
+      "required_token_positions_per_second": 76655.10846922097,
+      "required_pages_per_second_upper_estimate": 0.005303178268257281
     },
     {
       "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": 9563.113881906413,
-      "required_pages_per_second_upper_estimate": 0.004003876081827657
+      "required_token_positions_per_second": 9581.888558652621,
+      "required_pages_per_second_upper_estimate": 0.004011736646921461
     },
     {
       "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": 19126.227763812825,
-      "required_pages_per_second_upper_estimate": 0.00398348885065715
+      "required_token_positions_per_second": 19163.777117305242,
+      "required_pages_per_second_upper_estimate": 0.0039913093907465785
     },
     {
       "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": 28689.34164571924,
-      "required_pages_per_second_upper_estimate": 0.0017198822255726303
+      "required_token_positions_per_second": 28745.665675957862,
+      "required_pages_per_second_upper_estimate": 0.0017232587651836212
     },
     {
       "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": 28689.34164571924,
-      "required_pages_per_second_upper_estimate": 0.0022304701218770558
+      "required_token_positions_per_second": 28745.665675957862,
+      "required_pages_per_second_upper_estimate": 0.0022348490674848815
     },
     {
       "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": 28689.34164571924,
-      "required_pages_per_second_upper_estimate": 0.367996070416224
+      "required_token_positions_per_second": 28745.665675957862,
+      "required_pages_per_second_upper_estimate": 0.36871853460009313
     }
   ],
   "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": 0.08495708125812484,
-    "producer_p95_seconds": 0.2440374170000723,
-    "gpu_update_seconds": 0.9637028392432763,
-    "producer_batches_per_second": 11.770649193581674,
-    "gpu_batches_per_second": 1.0376642666999147,
-    "producer_headroom": 11.34340804759124,
+    "producer_batch_seconds": 0.13140172243737425,
+    "producer_p95_seconds": 0.23338307100129896,
+    "gpu_update_seconds": 0.9618145675131844,
+    "producer_batches_per_second": 7.61025031826807,
+    "gpu_batches_per_second": 1.0397014495065777,
+    "producer_headroom": 7.319649618532078,
     "sustainable": true,
-    "buffer_seconds": 60.713278872326406,
+    "buffer_seconds": 60.594317753330614,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366268.6460408284,
-    "upload_bytes_per_second": 4816634.339576173,
+    "generation_bytes_per_second": 374412.07998375414,
+    "upload_bytes_per_second": 4795191.198816356,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.59272514265669,
+    "estimated_full_shards_per_hour": 80.34011649736851,
     "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": 1.1643560489992524,
-      "bytes": 6319425,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 2.530195315999663,
       "bytes": 2235467,
@@ -562,6 +556,12 @@
       "bytes": 7741032,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.5200233909999952,
+      "bytes": 6160062,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -200,3 +200,4 @@
 {"step": 417100, "loss": 0.5619052052497864, "lr": 0.0003, "grad_norm": 0.16201801202103625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61052323072, "tokens_per_second": 152051.67114879543, "data_wait_s": 7.458998879883438e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 417200, "loss": 0.5778757333755493, "lr": 0.0003, "grad_norm": 0.19271627092949623, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61067065217, "tokens_per_second": 152987.1359427244, "data_wait_s": 1.6412999684689566e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 417300, "loss": 0.5162276029586792, "lr": 0.0003, "grad_norm": 0.15884595246507877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61081808801, "tokens_per_second": 153523.88054505913, "data_wait_s": 1.1416999768698588e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 417400, "loss": 0.5608070492744446, "lr": 0.0003, "grad_norm": 0.1427651690684036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61096659417, "tokens_per_second": 153616.18080028356, "data_wait_s": 1.3738001143792644e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:59beae7f163e6497f1c48c5fb53fe5c954a6fe5fa9c970d4599fad49773bf4de
-size 431133
+oid sha256:5ed63c74d58660fc93723fac17deb3b116f5ea80becf4aebb48d81da9921a6e8
+size 433251