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": 76914.9593431988,
-      "required_pages_per_second_upper_estimate": 0.005321155354656164
+      "required_token_positions_per_second": 77459.36985729232,
+      "required_pages_per_second_upper_estimate": 0.005358818937227582
     },
     {
       "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": 9614.36991789985,
-      "required_pages_per_second_upper_estimate": 0.004025335913750364
+      "required_token_positions_per_second": 9682.42123216154,
+      "required_pages_per_second_upper_estimate": 0.004053827577958696
     },
     {
       "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": 19228.7398357997,
-      "required_pages_per_second_upper_estimate": 0.00400483941182688
+      "required_token_positions_per_second": 19364.84246432308,
+      "required_pages_per_second_upper_estimate": 0.004033186000080618
     },
     {
       "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": 28843.109753699555,
-      "required_pages_per_second_upper_estimate": 0.001729100388855744
+      "required_token_positions_per_second": 29047.263696484617,
+      "required_pages_per_second_upper_estimate": 0.0017413391060006873
     },
     {
       "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": 28843.109753699555,
-      "required_pages_per_second_upper_estimate": 0.0022424249159180972
+      "required_token_positions_per_second": 29047.263696484617,
+      "required_pages_per_second_upper_estimate": 0.0022582969869913427
     },
     {
       "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": 28843.109753699555,
-      "required_pages_per_second_upper_estimate": 0.3699684426020646
+      "required_token_positions_per_second": 29047.263696484617,
+      "required_pages_per_second_upper_estimate": 0.372587110176686
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.4034197957030159,
-    "producer_p95_seconds": 0.19079471599980025,
-    "gpu_update_seconds": 0.9585651559798866,
-    "producer_batches_per_second": 2.478807462230154,
-    "gpu_batches_per_second": 1.043225902549897,
-    "producer_headroom": 2.3760984616767544,
+    "producer_batch_seconds": 0.5366441886093725,
+    "producer_p95_seconds": 0.2055995509999775,
+    "gpu_update_seconds": 0.9518280375354612,
+    "producer_batches_per_second": 1.8634320863351561,
+    "gpu_batches_per_second": 1.0506099427258615,
+    "producer_headroom": 1.7736669058170018,
     "sustainable": true,
-    "buffer_seconds": 60.38960482673286,
+    "buffer_seconds": 60.916994402269516,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 244346.57043727033,
-    "upload_bytes_per_second": 4825557.046242925,
+    "generation_bytes_per_second": 224663.9555499879,
+    "upload_bytes_per_second": 4818923.113755712,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.431085918794466,
+    "estimated_full_shards_per_hour": 48.207654951808244,
     "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": 1.4799904070000593,
-      "bytes": 10661870,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.9814531229999375,
       "bytes": 6026004,
@@ -562,6 +556,12 @@
       "bytes": 2656311,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.9071076060008636,
+      "bytes": 12136915,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -101,3 +101,4 @@
 {"step": 852600, "loss": 0.5513781309127808, "lr": 3e-05, "grad_norm": 0.3217502851203123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124796715130, "tokens_per_second": 124554.46547856432, "data_wait_s": 1.6227999367401935e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 852700, "loss": 0.5552854537963867, "lr": 3e-05, "grad_norm": 0.3149392184376773, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124811340335, "tokens_per_second": 149904.9143558171, "data_wait_s": 1.1418000212870538e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 852800, "loss": 0.5362980365753174, "lr": 3e-05, "grad_norm": 0.3573354062209486, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124826055684, "tokens_per_second": 142715.7767014391, "data_wait_s": 1.5141000403673388e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 852900, "loss": 0.5480002164840698, "lr": 3e-05, "grad_norm": 0.31473184307219726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124840690045, "tokens_per_second": 152391.08213359266, "data_wait_s": 1.203700048790779e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:894243fd9951e887afd730f60273c0014954285002dc07a411f150b6ca75e537
-size 219885
+oid sha256:55cf3c56016153f074f7839c38bd23baf0f61e54be29d0c27faf93f419153337
+size 222003