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": 77632.7225592504,
-      "required_pages_per_second_upper_estimate": 0.005370811879382728
+      "required_token_positions_per_second": 77476.08549637788,
+      "required_pages_per_second_upper_estimate": 0.005359975363925148
     },
     {
       "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": 9704.0903199063,
-      "required_pages_per_second_upper_estimate": 0.004062899972495419
+      "required_token_positions_per_second": 9684.510687047235,
+      "required_pages_per_second_upper_estimate": 0.0040547023890865575
     },
     {
       "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": 19408.1806398126,
-      "required_pages_per_second_upper_estimate": 0.004042212199130541
+      "required_token_positions_per_second": 19369.02137409447,
+      "required_pages_per_second_upper_estimate": 0.004034056356780738
     },
     {
       "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": 29112.2709597189,
-      "required_pages_per_second_upper_estimate": 0.0017452361921712388
+      "required_token_positions_per_second": 29053.532061141705,
+      "required_pages_per_second_upper_estimate": 0.0017417148848906411
     },
     {
       "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": 29112.2709597189,
-      "required_pages_per_second_upper_estimate": 0.002263351015771076
+      "required_token_positions_per_second": 29053.532061141705,
+      "required_pages_per_second_upper_estimate": 0.002258784324771809
     },
     {
       "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": 29112.2709597189,
-      "required_pages_per_second_upper_estimate": 0.37342095355009425
+      "required_token_positions_per_second": 29053.532061141705,
+      "required_pages_per_second_upper_estimate": 0.37266751402806153
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08428969727344793,
-    "producer_p95_seconds": 0.1979023589992721,
-    "gpu_update_seconds": 0.9497026198421644,
-    "producer_batches_per_second": 11.863846144278533,
-    "gpu_batches_per_second": 1.0529611892259438,
-    "producer_headroom": 11.267125764625684,
+    "producer_batch_seconds": 0.07981417509380151,
+    "producer_p95_seconds": 0.19079300400153443,
+    "gpu_update_seconds": 0.951622678503122,
+    "producer_batches_per_second": 12.529102741771762,
+    "gpu_batches_per_second": 1.0508366630910628,
+    "producer_headroom": 11.922978310365654,
     "sustainable": true,
-    "buffer_seconds": 60.78096766989852,
+    "buffer_seconds": 60.90385142419981,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 545048.9752310832,
-    "upload_bytes_per_second": 5799205.352431072,
+    "generation_bytes_per_second": 555665.2401398226,
+    "upload_bytes_per_second": 5779482.881375651,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 116.95482199382184,
+    "estimated_full_shards_per_hour": 119.2328253092385,
     "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": 1.0787890209994657,
-      "bytes": 6174644,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.431994874999873,
       "bytes": 7632897,
@@ -562,6 +556,12 @@
       "bytes": 7613200,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.303638638000848,
+      "bytes": 6027470,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -112,3 +112,4 @@
 {"step": 536200, "loss": 0.5933921933174133, "lr": 0.0003, "grad_norm": 0.1613617432365788, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78482185020, "tokens_per_second": 152032.92833788562, "data_wait_s": 1.3897999451728538e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 536300, "loss": 0.5682471394538879, "lr": 0.0003, "grad_norm": 0.1719186843746542, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78496839204, "tokens_per_second": 153581.7304489191, "data_wait_s": 9.982999472413212e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 536400, "loss": 0.5694395899772644, "lr": 0.0003, "grad_norm": 0.17182849504903272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78511539349, "tokens_per_second": 153951.41072311337, "data_wait_s": 1.1206000635866076e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 536500, "loss": 0.5985449552536011, "lr": 0.0003, "grad_norm": 0.1573570286032401, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78526239509, "tokens_per_second": 153347.29160659207, "data_wait_s": 1.7140999261755496e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cdfb5a55deca0121eb5a3193f2d1746de0d532093c01e9260f3dc4af502fec59
-size 243183
+oid sha256:7142148719e548cbfba627f104acc0ee134095ee46a119d47b9ee2465df3a375
+size 245301