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": 77027.92575684798,
-      "required_pages_per_second_upper_estimate": 0.005328970633270607
+      "required_token_positions_per_second": 77378.1174463459,
+      "required_pages_per_second_upper_estimate": 0.005353197706906741
     },
     {
       "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": 9628.490719605998,
-      "required_pages_per_second_upper_estimate": 0.004031247998548857
+      "required_token_positions_per_second": 9672.264680793238,
+      "required_pages_per_second_upper_estimate": 0.004049575241993699
     },
     {
       "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": 19256.981439211995,
-      "required_pages_per_second_upper_estimate": 0.004010721393036495
+      "required_token_positions_per_second": 19344.529361586476,
+      "required_pages_per_second_upper_estimate": 0.004028955316473127
     },
     {
       "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": 28885.472158817996,
-      "required_pages_per_second_upper_estimate": 0.0017316399503589472
+      "required_token_positions_per_second": 29016.79404237971,
+      "required_pages_per_second_upper_estimate": 0.0017395124967615663
     },
     {
       "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": 28885.472158817996,
-      "required_pages_per_second_upper_estimate": 0.0022457184065835284
+      "required_token_positions_per_second": 29016.79404237971,
+      "required_pages_per_second_upper_estimate": 0.002255928105406531
     },
     {
       "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": 28885.472158817996,
-      "required_pages_per_second_upper_estimate": 0.3705118220497171
+      "required_token_positions_per_second": 29016.79404237971,
+      "required_pages_per_second_upper_estimate": 0.37219627816959394
     }
   ],
   "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": 0.06710805025008426,
-    "producer_p95_seconds": 0.2398295450002479,
-    "gpu_update_seconds": 0.9571593584479378,
-    "producer_batches_per_second": 14.90134188481723,
-    "gpu_batches_per_second": 1.044758107596137,
-    "producer_headroom": 14.262958838485044,
+    "producer_batch_seconds": 0.4954559147342934,
+    "producer_p95_seconds": 0.2526271220012859,
+    "gpu_update_seconds": 0.9528275232480695,
+    "producer_batches_per_second": 2.018343045791041,
+    "gpu_batches_per_second": 1.049507886370794,
+    "producer_headroom": 1.9231328053860424,
     "sustainable": true,
-    "buffer_seconds": 60.30103958222008,
+    "buffer_seconds": 60.02813396462838,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379329.14512087335,
-    "upload_bytes_per_second": 4831414.163810238,
+    "generation_bytes_per_second": 333008.77131654706,
+    "upload_bytes_per_second": 4871090.117931575,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.39520421237611,
+    "estimated_full_shards_per_hour": 71.45593027708348,
     "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": 1.127117382000506,
-      "bytes": 7535662,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.3609303779994661,
       "bytes": 10666543,
@@ -562,6 +556,12 @@
       "bytes": 7759537,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.2661478619993431,
+      "bytes": 11756535,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -236,3 +236,4 @@
 {"step": 948600, "loss": 0.5446297526359558, "lr": 3e-05, "grad_norm": 0.3369444155076515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138848824027, "tokens_per_second": 153227.1068928778, "data_wait_s": 1.4082997950026765e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 948700, "loss": 0.5530696511268616, "lr": 3e-05, "grad_norm": 0.3431264844313009, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138863500206, "tokens_per_second": 153369.66960290697, "data_wait_s": 1.3081000361125916e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 948800, "loss": 0.4947313070297241, "lr": 3e-05, "grad_norm": 0.3324267838111315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138878112624, "tokens_per_second": 152504.77627283018, "data_wait_s": 1.4070999895920977e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 948900, "loss": 0.5397599339485168, "lr": 3e-05, "grad_norm": 0.35230049628193194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138892548315, "tokens_per_second": 150450.3322366843, "data_wait_s": 1.689300188445486e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:95f2d085f2a4ddcade56e7dd6351e411f34ca5a5282a45df29ef76324b8ec56d
-size 507381
+oid sha256:11b3819b3ef8d6b98d9f7ab24d98d77a09b311c8d1fca137588ded6948f9e699
+size 509499