Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75303.1039230757,
-      "required_pages_per_second_upper_estimate": 0.006527082194080181
+      "required_token_positions_per_second": 75831.8280650318,
+      "required_pages_per_second_upper_estimate": 0.006572910662665849
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9412.887990384463,
-      "required_pages_per_second_upper_estimate": 0.003754941754581324
+      "required_token_positions_per_second": 9478.978508128976,
+      "required_pages_per_second_upper_estimate": 0.0037813062502509076
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18825.775980768925,
-      "required_pages_per_second_upper_estimate": 0.004355011950809692
+      "required_token_positions_per_second": 18957.95701625795,
+      "required_pages_per_second_upper_estimate": 0.004385589707063307
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28238.66397115339,
-      "required_pages_per_second_upper_estimate": 0.0017195864790235471
+      "required_token_positions_per_second": 28436.93552438693,
+      "required_pages_per_second_upper_estimate": 0.0017316601763650247
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28238.66397115339,
-      "required_pages_per_second_upper_estimate": 0.0021738905221243385
+      "required_token_positions_per_second": 28436.93552438693,
+      "required_pages_per_second_upper_estimate": 0.002189154015143044
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28238.66397115339,
-      "required_pages_per_second_upper_estimate": 0.350629697792982
+      "required_token_positions_per_second": 28436.93552438693,
+      "required_pages_per_second_upper_estimate": 0.3530915669119402
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.056825645749960074,
-    "producer_p95_seconds": 0.19185754299996916,
-    "gpu_update_seconds": 0.9790831474266357,
-    "producer_batches_per_second": 17.59768827617243,
-    "gpu_batches_per_second": 1.0213637142344252,
-    "producer_headroom": 17.229600024867707,
+    "producer_batch_seconds": 0.1059172976172178,
+    "producer_p95_seconds": 0.2050307800000155,
+    "gpu_update_seconds": 0.9722566616325323,
+    "producer_batches_per_second": 9.441328493991344,
+    "gpu_batches_per_second": 1.0285349943716338,
+    "producer_headroom": 9.17939452294413,
     "sustainable": true,
-    "buffer_seconds": 60.703155140451415,
+    "buffer_seconds": 60.279913021217006,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 420756.39741989586,
+    "generation_bytes_per_second": 422528.95386137476,
     "upload_bytes_per_second": 5690772.691029994,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.28452817866952,
+    "estimated_full_shards_per_hour": 90.66487752800877,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -76,3 +76,4 @@
 {"step": 362600, "loss": 0.5897896364331245, "lr": 0.0003, "grad_norm": 0.15633946977455013, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53074429274, "tokens_per_second": 146945.11608346397, "data_wait_s": 1.0820000170497224e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 362700, "loss": 0.6040910072624683, "lr": 0.0003, "grad_norm": 0.15407801479255753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53088949404, "tokens_per_second": 147469.91447182413, "data_wait_s": 1.1153999366797507e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 362800, "loss": 0.5554407816380262, "lr": 0.0003, "grad_norm": 0.21308853438439007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53103702390, "tokens_per_second": 151021.7899532143, "data_wait_s": 1.0656998711056076e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 362900, "loss": 0.6023930348455906, "lr": 0.0003, "grad_norm": 0.16299847025574002, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53118379197, "tokens_per_second": 150001.53544897021, "data_wait_s": 1.1322999853291549e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6a6b6ac9c5cfde3e37fe1210944ca751597fc791c4efa6aa260b5ca295ca0ae1
-size 166152
+oid sha256:abe625e89bbd413f4533ad7e909c15a3e0ef0a40b89d496d44180df4bc856ff8
+size 168270