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": 76101.06120642765,
-      "required_pages_per_second_upper_estimate": 0.005264847993050473
+      "required_token_positions_per_second": 75706.48897661196,
+      "required_pages_per_second_upper_estimate": 0.005237550570658115
     },
     {
       "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": 9512.632650803456,
-      "required_pages_per_second_upper_estimate": 0.0039827406445328275
+      "required_token_positions_per_second": 9463.311122076495,
+      "required_pages_per_second_upper_estimate": 0.003962090750405488
     },
     {
       "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": 19025.265301606913,
-      "required_pages_per_second_upper_estimate": 0.003962461032338836
+      "required_token_positions_per_second": 18926.62224415299,
+      "required_pages_per_second_upper_estimate": 0.003941916284863545
     },
     {
       "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": 28537.897952410367,
-      "required_pages_per_second_upper_estimate": 0.0017108034073998937
+      "required_token_positions_per_second": 28389.93336622949,
+      "required_pages_per_second_upper_estimate": 0.0017019331563871915
     },
     {
       "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": 28537.897952410367,
-      "required_pages_per_second_upper_estimate": 0.002218696040852699
+      "required_token_positions_per_second": 28389.93336622949,
+      "required_pages_per_second_upper_estimate": 0.0022071924451045684
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28537.897952410367,
-      "required_pages_per_second_upper_estimate": 0.3660535133260267
+      "required_token_positions_per_second": 28389.93336622949,
+      "required_pages_per_second_upper_estimate": 0.3641555824864931
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.07863502711716563,
-    "producer_p95_seconds": 0.13925462099996366,
-    "gpu_update_seconds": 0.9688169761524007,
-    "producer_batches_per_second": 12.716979146074523,
-    "gpu_batches_per_second": 1.0321867025611389,
-    "producer_headroom": 12.320425282093058,
+    "producer_batch_seconds": 0.3206198619452927,
+    "producer_p95_seconds": 0.15215030499985005,
+    "gpu_update_seconds": 0.973866322380593,
+    "producer_batches_per_second": 3.118958363754239,
+    "gpu_batches_per_second": 1.0268349741836476,
+    "producer_headroom": 3.0374485113675322,
     "sustainable": true,
-    "buffer_seconds": 60.066652521448844,
+    "buffer_seconds": 60.379711987596764,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 68,
+    "upload_pages_per_commit": 67,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318600.83028303465,
-    "upload_bytes_per_second": 4268655.294811884,
+    "generation_bytes_per_second": 300149.4584818252,
+    "upload_bytes_per_second": 3695594.3210382494,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.36432153099327,
+    "estimated_full_shards_per_hour": 64.40508666840617,
     "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": [
@@ -190,6 +190,12 @@
       "bytes": 11915224,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.1385679780000828,
+      "bytes": 2605060,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791033297667653750-ba4975eb/train.jsonl

@@ -13,3 +13,4 @@
 {"step": 601300, "loss": 0.5702246427536011, "lr": 0.0003, "grad_norm": 0.16696146154166297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88012437486, "tokens_per_second": 151797.32403943306, "data_wait_s": 1.4603999943574308e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 601400, "loss": 0.6435558795928955, "lr": 0.0003, "grad_norm": 0.18562189922226408, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88027174557, "tokens_per_second": 153222.16980978753, "data_wait_s": 2.5678000383777544e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 601500, "loss": 0.5248953104019165, "lr": 0.0003, "grad_norm": 0.15204377231219687, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88041812100, "tokens_per_second": 152086.48631346013, "data_wait_s": 1.436200000171084e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 601600, "loss": 0.5386746525764465, "lr": 0.0003, "grad_norm": 0.1795039055292808, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88056375401, "tokens_per_second": 150902.21547646535, "data_wait_s": 1.4191999980539549e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791033297.83ac8725-d818-41af-bbb1-78dcd88792cf-7zwhs.737.0.1791033297667653750-ba4975eb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:786ac64e776650d4835908962ccef1cddcc7352851ca932773c8a320ce7b4238
-size 31935
+oid sha256:cf1321273adbcf83c2212dcf16d25d312059f401ce3be8feaa7da8324d76c026
+size 34053