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": 77048.52146353533,
-      "required_pages_per_second_upper_estimate": 0.005330395492047862
+      "required_token_positions_per_second": 76340.2271798615,
+      "required_pages_per_second_upper_estimate": 0.005281394050034141
     },
     {
       "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": 9631.065182941917,
-      "required_pages_per_second_upper_estimate": 0.004032325872586699
+      "required_token_positions_per_second": 9542.528397482687,
+      "required_pages_per_second_upper_estimate": 0.003995257369373241
     },
     {
       "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": 19262.130365883833,
-      "required_pages_per_second_upper_estimate": 0.0040117937786759116
+      "required_token_positions_per_second": 19085.056794965374,
+      "required_pages_per_second_upper_estimate": 0.0039749140235976884
     },
     {
       "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": 28893.19554882575,
-      "required_pages_per_second_upper_estimate": 0.0017321029557969786
+      "required_token_positions_per_second": 28627.58519244806,
+      "required_pages_per_second_upper_estimate": 0.001716180020495667
     },
     {
       "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": 28893.19554882575,
-      "required_pages_per_second_upper_estimate": 0.002246318866185029
+      "required_token_positions_per_second": 28627.58519244806,
+      "required_pages_per_second_upper_estimate": 0.0022256688292731515
     },
     {
       "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": 28893.19554882575,
-      "required_pages_per_second_upper_estimate": 0.3706108894040065
+      "required_token_positions_per_second": 28627.58519244806,
+      "required_pages_per_second_upper_estimate": 0.3672039249425746
     }
   ],
   "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": 0.05688460075785429,
-    "producer_p95_seconds": 0.1836397549996036,
-    "gpu_update_seconds": 0.9569035018393334,
-    "producer_batches_per_second": 17.579450091542146,
-    "gpu_batches_per_second": 1.0450374547463017,
-    "producer_headroom": 16.82183735300647,
+    "producer_batch_seconds": 0.10123071792970961,
+    "producer_p95_seconds": 0.21726470599969616,
+    "gpu_update_seconds": 0.9657817735634064,
+    "producer_batches_per_second": 9.878424459010143,
+    "gpu_batches_per_second": 1.0354305986851873,
+    "producer_headroom": 9.54040229403495,
     "sustainable": true,
-    "buffer_seconds": 60.284920615878,
+    "buffer_seconds": 60.844251734494605,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 469989.2778609476,
-    "upload_bytes_per_second": 5026063.997653216,
+    "generation_bytes_per_second": 469484.45382974844,
+    "upload_bytes_per_second": 5027427.0942602055,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.84875823851891,
+    "estimated_full_shards_per_hour": 100.74043475312557,
     "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": 0.9787746179999885,
-      "bytes": 1132423,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 0.8866051049999442,
       "bytes": 7112636,
@@ -562,6 +556,12 @@
       "bytes": 16421464,
       "pages": 42,
       "shards": 4
+    },
+    {
+      "seconds": 0.9113021489993116,
+      "bytes": 908789,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -90,3 +90,4 @@
 {"step": 406100, "loss": 0.5753645300865173, "lr": 0.0003, "grad_norm": 0.16737049978570817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59442364895, "tokens_per_second": 152202.8521291031, "data_wait_s": 1.2889999197795987e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 406200, "loss": 0.6132565140724182, "lr": 0.0003, "grad_norm": 0.1855223052537949, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59457056283, "tokens_per_second": 152301.5334593381, "data_wait_s": 1.5722000171081163e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 406300, "loss": 0.6408392786979675, "lr": 0.0003, "grad_norm": 0.18752841436474124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59471756710, "tokens_per_second": 152487.0561394124, "data_wait_s": 9.763001799001358e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 406400, "loss": 0.5909572243690491, "lr": 0.0003, "grad_norm": 0.1529850649040701, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59486498541, "tokens_per_second": 153533.83831338357, "data_wait_s": 1.0314999599358998e-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 sha256:99685e14e1ef367d98429eca10b53b7c387d1ebeca58ff39c37b77d9c87c29e4
-size 196587
+oid sha256:f017bcdedd561de377d33377e89442ae175d08aeab979ee6e96d11712840eb62
+size 198705