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": 76344.01582692025,
-      "required_pages_per_second_upper_estimate": 0.005281656157428538
+      "required_token_positions_per_second": 76536.75638231971,
+      "required_pages_per_second_upper_estimate": 0.005294990396270776
     },
     {
       "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": 9543.001978365031,
-      "required_pages_per_second_upper_estimate": 0.003995455647799184
+      "required_token_positions_per_second": 9567.094547789964,
+      "required_pages_per_second_upper_estimate": 0.0040055427035073435
     },
     {
       "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": 19086.003956730063,
-      "required_pages_per_second_upper_estimate": 0.003975111292414938
+      "required_token_positions_per_second": 19134.189095579928,
+      "required_pages_per_second_upper_estimate": 0.0039851469861098
     },
     {
       "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": 28629.005935095094,
-      "required_pages_per_second_upper_estimate": 0.0017162651918480084
+      "required_token_positions_per_second": 28701.28364336989,
+      "required_pages_per_second_upper_estimate": 0.001720598130097412
     },
     {
       "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": 28629.005935095094,
-      "required_pages_per_second_upper_estimate": 0.0022257792857647723
+      "required_token_positions_per_second": 28701.28364336989,
+      "required_pages_per_second_upper_estimate": 0.0022313985596670973
     },
     {
       "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": 28629.005935095094,
-      "required_pages_per_second_upper_estimate": 0.3672221487037762
+      "required_token_positions_per_second": 28701.28364336989,
+      "required_pages_per_second_upper_estimate": 0.36814924953976846
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11139093747618745,
-    "producer_p95_seconds": 0.31134687300072983,
-    "gpu_update_seconds": 0.965733845690656,
-    "producer_batches_per_second": 8.97739100376792,
-    "gpu_batches_per_second": 1.035481985499678,
-    "producer_headroom": 8.669770338337491,
+    "producer_batch_seconds": 0.08796119544524572,
+    "producer_p95_seconds": 0.30559524000273086,
+    "gpu_update_seconds": 0.9633018628554196,
+    "producer_batches_per_second": 11.368649492974232,
+    "gpu_batches_per_second": 1.038096196591793,
+    "producer_headroom": 10.951441234732398,
     "sustainable": true,
-    "buffer_seconds": 60.84123227851133,
+    "buffer_seconds": 60.68801735989144,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 280420.5868789041,
-    "upload_bytes_per_second": 4845392.632499201,
+    "generation_bytes_per_second": 299952.7339558699,
+    "upload_bytes_per_second": 4756086.631623678,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.17173008704511,
+    "estimated_full_shards_per_hour": 64.36287416464876,
     "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": 1.339776704000542,
-      "bytes": 13443293,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 1.7319635569983802,
       "bytes": 934829,
@@ -562,6 +556,12 @@
       "bytes": 7282459,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.3900761400000192,
+      "bytes": 5687350,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -385,3 +385,4 @@
 {"step": 758500, "loss": 0.5073688626289368, "lr": 5e-05, "grad_norm": 0.2388039053231994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111022668129, "tokens_per_second": 150572.55613309293, "data_wait_s": 1.0691997886169702e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 758600, "loss": 0.593370258808136, "lr": 5e-05, "grad_norm": 0.3061563083484681, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111037212132, "tokens_per_second": 150036.55847191025, "data_wait_s": 1.4796998584643006e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 758700, "loss": 0.5871933102607727, "lr": 5e-05, "grad_norm": 0.2487585261294385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111051956469, "tokens_per_second": 153047.42144757885, "data_wait_s": 1.1112999345641583e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 758800, "loss": 0.5338184237480164, "lr": 5e-05, "grad_norm": 0.2515999769671483, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111066630057, "tokens_per_second": 151498.08155184865, "data_wait_s": 1.0244999430142343e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6d9e5c6ae721ddc85990841a2cf899c9d6de42c427c087abca81ac05dd40e8bc
-size 825312
+oid sha256:2524b2cea8afa78199a926c2ce412dc9443d3041485bb9077faef17fa322c5a9
+size 827430