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": 77214.94122731619,
-      "required_pages_per_second_upper_estimate": 0.005341908797453273
+      "required_token_positions_per_second": 77097.92411569778,
+      "required_pages_per_second_upper_estimate": 0.0053338132821544055
     },
     {
       "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": 9651.867653414523,
-      "required_pages_per_second_upper_estimate": 0.004041035432568597
+      "required_token_positions_per_second": 9637.240514462223,
+      "required_pages_per_second_upper_estimate": 0.004034911354938665
     },
     {
       "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": 19303.735306829047,
-      "required_pages_per_second_upper_estimate": 0.004020458990678723
+      "required_token_positions_per_second": 19274.481028924445,
+      "required_pages_per_second_upper_estimate": 0.004014366096078383
     },
     {
       "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": 28955.60296024357,
-      "required_pages_per_second_upper_estimate": 0.001735844184820877
+      "required_token_positions_per_second": 28911.72154338667,
+      "required_pages_per_second_upper_estimate": 0.0017332135608833497
     },
     {
       "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": 28955.60296024357,
-      "required_pages_per_second_upper_estimate": 0.0022511707679215716
+      "required_token_positions_per_second": 28911.72154338667,
+      "required_pages_per_second_upper_estimate": 0.002247759180774902
     },
     {
       "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": 28955.60296024357,
-      "required_pages_per_second_upper_estimate": 0.3714113846698166
+      "required_token_positions_per_second": 28911.72154338667,
+      "required_pages_per_second_upper_estimate": 0.3708485209705708
     }
   ],
   "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": 0.08379778353116762,
-    "producer_p95_seconds": 0.18346639999981562,
-    "gpu_update_seconds": 0.9548411075383605,
-    "producer_batches_per_second": 11.933489859288002,
-    "gpu_batches_per_second": 1.04729466725418,
-    "producer_headroom": 11.39458667404035,
+    "producer_batch_seconds": 0.07486276451572849,
+    "producer_p95_seconds": 0.19115245300054085,
+    "gpu_update_seconds": 0.9562903391453099,
+    "producer_batches_per_second": 13.35777547715196,
+    "gpu_batches_per_second": 1.0457075211005016,
+    "producer_headroom": 12.773911641272553,
     "sustainable": true,
-    "buffer_seconds": 60.154989774916714,
+    "buffer_seconds": 60.24629136615452,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299856.77172942745,
-    "upload_bytes_per_second": 4884652.560857204,
+    "generation_bytes_per_second": 326583.97403639957,
+    "upload_bytes_per_second": 4908990.026384033,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.3422829047405,
+    "estimated_full_shards_per_hour": 70.07731834238997,
     "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": 1.509009006000042,
-      "bytes": 6599366,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.1468410420000055,
       "bytes": 7711330,
@@ -562,6 +556,12 @@
       "bytes": 7697898,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.9798488249998627,
+      "bytes": 6153509,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -90,3 +90,4 @@
 {"step": 851500, "loss": 0.5021769404411316, "lr": 3e-05, "grad_norm": 0.29591102837388067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124635617791, "tokens_per_second": 151674.04648592143, "data_wait_s": 1.0235000445391051e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 851600, "loss": 0.5318317413330078, "lr": 3e-05, "grad_norm": 0.3137795975406802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124650436349, "tokens_per_second": 154114.831035925, "data_wait_s": 1.3015000149607658e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 851700, "loss": 0.4832938611507416, "lr": 3e-05, "grad_norm": 0.3133136192186214, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124665061893, "tokens_per_second": 152238.17009867422, "data_wait_s": 8.699000318301842e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 851800, "loss": 0.5552166104316711, "lr": 3e-05, "grad_norm": 0.31155580857427795, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124679698959, "tokens_per_second": 138299.4433037356, "data_wait_s": 1.0236999514745548e-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 sha256:13edec4b6d064c69bd4851c3bac8abe5f04c2ef4cb924e8b47daa4dabf5137ff
-size 196587
+oid sha256:ee5ec1f49d907fb629277f32e8cb2222e011bd2072b6fae70715903ec9c841a4
+size 198705