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": 77352.36512836233,
-      "required_pages_per_second_upper_estimate": 0.005351416101795028
+      "required_token_positions_per_second": 76953.00268853996,
+      "required_pages_per_second_upper_estimate": 0.005323787281559588
     },
     {
       "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": 9669.045641045292,
-      "required_pages_per_second_upper_estimate": 0.004048227497272428
+      "required_token_positions_per_second": 9619.125336067495,
+      "required_pages_per_second_upper_estimate": 0.0040273269080327335
     },
     {
       "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": 19338.091282090583,
-      "required_pages_per_second_upper_estimate": 0.004027614434297706
+      "required_token_positions_per_second": 19238.25067213499,
+      "required_pages_per_second_upper_estimate": 0.0040068202682179005
     },
     {
       "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": 29007.136923135873,
-      "required_pages_per_second_upper_estimate": 0.0017389335672084653
+      "required_token_positions_per_second": 28857.376008202486,
+      "required_pages_per_second_upper_estimate": 0.0017299556290298853
     },
     {
       "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": 29007.136923135873,
-      "required_pages_per_second_upper_estimate": 0.0022551773068625013
+      "required_token_positions_per_second": 28857.376008202486,
+      "required_pages_per_second_upper_estimate": 0.002243534054455078
     },
     {
       "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": 29007.136923135873,
-      "required_pages_per_second_upper_estimate": 0.37207240701294075
+      "required_token_positions_per_second": 28857.376008202486,
+      "required_pages_per_second_upper_estimate": 0.3701514347969175
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.40391197483597807,
-    "producer_p95_seconds": 0.25802149399896734,
-    "gpu_update_seconds": 0.9531447406637421,
-    "producer_batches_per_second": 2.4757869592900366,
-    "gpu_batches_per_second": 1.049158598203699,
-    "producer_headroom": 2.359783319251177,
+    "producer_batch_seconds": 0.12996821995281493,
+    "producer_p95_seconds": 0.26627065499997116,
+    "gpu_update_seconds": 0.9580912690100885,
+    "producer_batches_per_second": 7.694188628289676,
+    "gpu_batches_per_second": 1.0437418984448237,
+    "producer_headroom": 7.371734946881047,
     "sustainable": true,
-    "buffer_seconds": 60.04811866181575,
+    "buffer_seconds": 60.35974994763557,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337631.6727375394,
+    "generation_bytes_per_second": 322930.6597256934,
     "upload_bytes_per_second": 3887759.570666917,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.44789730639111,
+    "estimated_full_shards_per_hour": 69.293402136117,
     "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-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -218,3 +218,4 @@
 {"step": 666800, "loss": 0.5390537977218628, "lr": 0.0001, "grad_norm": 0.1958525864032979, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97602555346, "tokens_per_second": 151428.08310205027, "data_wait_s": 9.765000868355855e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 666900, "loss": 0.5449493527412415, "lr": 0.0001, "grad_norm": 0.18722814729427356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97617364320, "tokens_per_second": 155077.09781617267, "data_wait_s": 1.1454998457338661e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 667000, "loss": 0.5595698356628418, "lr": 0.0001, "grad_norm": 0.18506470624595225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97632034622, "tokens_per_second": 152100.2274270048, "data_wait_s": 1.0599000233924016e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 667100, "loss": 0.5123287439346313, "lr": 0.0001, "grad_norm": 0.19962370620755204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97646820828, "tokens_per_second": 153365.6932093714, "data_wait_s": 1.3494001905201003e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:89b53a197b3e3b9b06c2114d269674517b760a49cf126378f2907e1ca2655595
-size 469257
+oid sha256:a0d2b11f88adbf7ab63b7a7f6af0ec3725804ec7098a588020117227f3fe8163
+size 471375