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": 75558.03277493644,
-      "required_pages_per_second_upper_estimate": 0.006549178780848201
+      "required_token_positions_per_second": 75623.98351108948,
+      "required_pages_per_second_upper_estimate": 0.006554895223507334
     },
     {
       "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": 9444.754096867055,
-      "required_pages_per_second_upper_estimate": 0.0037676536208979794
+      "required_token_positions_per_second": 9452.997938886185,
+      "required_pages_per_second_upper_estimate": 0.0037709422127358326
     },
     {
       "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": 18889.50819373411,
-      "required_pages_per_second_upper_estimate": 0.004369755276630549
+      "required_token_positions_per_second": 18905.99587777237,
+      "required_pages_per_second_upper_estimate": 0.004373569412159471
     },
     {
       "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": 28334.262290601164,
-      "required_pages_per_second_upper_estimate": 0.0017254079151122976
+      "required_token_positions_per_second": 28358.993816658553,
+      "required_pages_per_second_upper_estimate": 0.0017269139352929033
     },
     {
       "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": 28334.262290601164,
-      "required_pages_per_second_upper_estimate": 0.0021812499453885137
+      "required_token_positions_per_second": 28358.993816658553,
+      "required_pages_per_second_upper_estimate": 0.0021831538467256576
     },
     {
       "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": 28334.262290601164,
-      "required_pages_per_second_upper_estimate": 0.35181670897352973
+      "required_token_positions_per_second": 28358.993816658553,
+      "required_pages_per_second_upper_estimate": 0.3521237917560693
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1122944519843827,
-    "producer_p95_seconds": 0.3846020819983096,
-    "gpu_update_seconds": 0.9757797720807855,
-    "producer_batches_per_second": 8.905159447584058,
-    "gpu_batches_per_second": 1.0248214080801925,
-    "producer_headroom": 8.689474456106625,
+    "producer_batch_seconds": 0.2577296311719124,
+    "producer_p95_seconds": 0.39432812300219666,
+    "gpu_update_seconds": 0.9749288066687012,
+    "producer_batches_per_second": 3.8800350408020177,
+    "gpu_batches_per_second": 1.025715922188171,
+    "producer_headroom": 3.7827579321618563,
     "sustainable": true,
-    "buffer_seconds": 60.4983458690087,
+    "buffer_seconds": 60.445586013459476,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 165,
+    "upload_pages_per_commit": 169,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 620246.8959293183,
-    "upload_bytes_per_second": 5628574.884011771,
+    "generation_bytes_per_second": 559402.2649098603,
+    "upload_bytes_per_second": 5721186.235684265,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.54527262883025,
+    "estimated_full_shards_per_hour": 60.01735191570214,
     "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": 0.9736092290004308,
-      "bytes": 2727259,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.154159282999899,
       "bytes": 5704653,
@@ -562,6 +556,12 @@
       "bytes": 1203686,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.0235948650006321,
+      "bytes": 9738901,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -335,3 +335,4 @@
 {"step": 228500, "loss": 0.6087172422558069, "lr": 0.0003, "grad_norm": 0.1542153343676074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33445805351, "tokens_per_second": 150633.85695596994, "data_wait_s": 9.520997991785407e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 228600, "loss": 0.5959756262600422, "lr": 0.0003, "grad_norm": 0.14663853985307887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33460544310, "tokens_per_second": 150725.96862467233, "data_wait_s": 1.0165000276174396e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 228700, "loss": 0.593542791903019, "lr": 0.0003, "grad_norm": 0.15870941771635896, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33475266402, "tokens_per_second": 149403.2379047505, "data_wait_s": 1.26349987112917e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 228800, "loss": 0.5507429484277964, "lr": 0.0003, "grad_norm": 0.14181171220754413, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33489760417, "tokens_per_second": 146354.72225646026, "data_wait_s": 9.732000762596726e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ce69f4929b776866a56ae89c405a973a39e4b9f79f36bee665580d9afaa62aeb
-size 718629
+oid sha256:615c89f5a4b69b353ddb62edf4530a52f3e183cd1f7b99ff0bd43cc183b6fe6a
+size 720747